tests/testthat/test-stPGOcc-NNGP.R
In doserjef/spOccupancy: Single-Species, Multi-Species, and Integrated Spatial Occupancy Models
# Test stPGOcc.R ---------------------------------------------------------
# NNGP --------------------------------------------------------------------
skip_on_cran()
# Intercept only ----------------------------------------------------------
set.seed(150)
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list()
alpha <- c(-1)
p.det <- length(alpha)
p.RE <- list()
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
# X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
# X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
# X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
# X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1])
det.covs <- list(int = X.p[, , , 1])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ 1
det.formula <- ~ 1
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
ar1 = TRUE,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check non-integer n.post -------------
test_that("non-integer n.post", {
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.thin = 13,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
n.omp.threads = 1,
verbose = FALSE))
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, FALSE)
expect_equal(out$psiRE, FALSE)
})
# Check RE error ----------------------
# test_that("random effect gives error when non-numeric", {
# data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
# data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
# data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# })
# Check RE levels ---------------------
# test_that("random effect levels are correct", {
# expect_equal(sort(unique(unlist(out$p.re.level.names))),
# sort(unique(c(X.p.re))))
# expect_equal(sort(unique(unlist(out$re.level.names))),
# sort(unique(c(X.re))))
# })
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
# tmp.data <- data.list
# tmp.data$occ.covs$trend[3, ] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
# tmp.data <- data.list
# tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
# X.0.full <- abind::abind(X.0, X.re.0, along = 3)
# dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend', 'occ.factor.1')
X.0.full <- X.0
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
# X.p.0 <- abind::abind(dat$X.p[, , 1, ], dat$X.p.re[, , 1, ], along = 3)
# dimnames(X.p.0)[[3]] <- c('(Intercept)', 'det.cov.1', 'det.factor.1', 'det.factor.2')
X.p.0 <- array(dat$X.p[, , 1, ], dim = c(J, n.time.max, p.det))
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Occurrence covariate only -----------------------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4, 0.8)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list()
alpha <- c(-1)
p.det <- length(alpha)
p.RE <- list()
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
# X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
# X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
# X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
# X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
trend = X[, , 2])
det.covs <- list(int = X.p[, , , 1])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ trend
det.formula <- ~ 1
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 1,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, FALSE)
expect_equal(out$psiRE, FALSE)
})
# Check RE error ----------------------
# test_that("random effect gives error when non-numeric", {
# data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
# data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
# data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# })
# Check RE levels ---------------------
# test_that("random effect levels are correct", {
# expect_equal(sort(unique(unlist(out$p.re.level.names))),
# sort(unique(c(X.p.re))))
# expect_equal(sort(unique(unlist(out$re.level.names))),
# sort(unique(c(X.re))))
# })
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
ar1 = TRUE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
tmp.data <- data.list
tmp.data$occ.covs$trend[3, ] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
# tmp.data <- data.list
# tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- X.0
# X.0.full <- abind::abind(X.0, X.re.0, along = 3)
dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
X.p.0 <- array(dat$X.p[, , 1, ], dim = c(J, n.time.max, 1))
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Detection covariate only ------------------------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list()
alpha <- c(-1, 0.5, -0.4)
p.det <- length(alpha)
p.RE <- list()
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
# X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
# X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
# X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
# X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1])
det.covs <- list(int = X.p[, , , 1],
det.cov.1 = X.p[, , , 2],
det.cov.2 = X.p[, , , 3])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ 1
det.formula <- ~ det.cov.1 + det.cov.2
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, FALSE)
expect_equal(out$psiRE, FALSE)
})
# Check RE error ----------------------
# test_that("random effect gives error when non-numeric", {
# data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
# data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
# data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# })
# Check RE levels ---------------------
# test_that("random effect levels are correct", {
# expect_equal(sort(unique(unlist(out$p.re.level.names))),
# sort(unique(c(X.p.re))))
# expect_equal(sort(unique(unlist(out$re.level.names))),
# sort(unique(c(X.re))))
# })
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
# tmp.data <- data.list
# tmp.data$occ.covs$trend[3, ] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- 1
tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- X.0
# X.0.full <- abind::abind(X.0, X.re.0, along = 3)
# dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
X.p.0 <- dat$X.p[, , 1, ]
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Covariates on both ------------------------------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4, 0.4)
p.occ <- length(beta)
trend <- FALSE
sp.only <- 0
psi.RE <- list()
alpha <- c(-1, 0.5, -0.4)
p.det <- length(alpha)
p.RE <- list()
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
# X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
# X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
# X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
# X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
occ.cov.1 = X[, , 2])
det.covs <- list(int = X.p[, , , 1],
det.cov.1 = X.p[, , , 2],
det.cov.2 = X.p[, , , 3])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ occ.cov.1
det.formula <- ~ det.cov.1 + det.cov.2
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, FALSE)
expect_equal(out$psiRE, FALSE)
})
# Check RE error ----------------------
# test_that("random effect gives error when non-numeric", {
# data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
# data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
# data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# })
# Check RE levels ---------------------
# test_that("random effect levels are correct", {
# expect_equal(sort(unique(unlist(out$p.re.level.names))),
# sort(unique(c(X.p.re))))
# expect_equal(sort(unique(unlist(out$re.level.names))),
# sort(unique(c(X.re))))
# })
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
tmp.data <- data.list
tmp.data$occ.covs$occ.cov.1[3, ] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- 1
tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- X.0
# X.0.full <- abind::abind(X.0, X.re.0, along = 3)
# dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
X.p.0 <- dat$X.p[, , 1, ]
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Interactions on both ----------------------------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4, 0.4, 0.2)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list()
alpha <- c(-1, 0.5, -0.4)
p.det <- length(alpha)
p.RE <- list()
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
# X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
# X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
# X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
# X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
trend = X[, , 2],
occ.cov.1 = X[, , 3])
det.covs <- list(int = X.p[, , , 1],
det.cov.1 = X.p[, , , 2],
det.cov.2 = X.p[, , , 3])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ trend * occ.cov.1
det.formula <- ~ det.cov.1 * det.cov.2
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
ar1 = TRUE,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, FALSE)
expect_equal(out$psiRE, FALSE)
})
# Check RE error ----------------------
# test_that("random effect gives error when non-numeric", {
# data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
# data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
# data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# })
# Check RE levels ---------------------
# test_that("random effect levels are correct", {
# expect_equal(sort(unique(unlist(out$p.re.level.names))),
# sort(unique(c(X.p.re))))
# expect_equal(sort(unique(unlist(out$re.level.names))),
# sort(unique(c(X.re))))
# })
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
tmp.data <- data.list
tmp.data$occ.covs$trend[3, ] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- 1
tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- abind(X.0, X.0[, , 3, drop = FALSE] * X.0[, , 2, drop = FALSE])
# X.0.full <- abind::abind(X.0, X.re.0, along = 3)
# dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
X.p.0 <- dat$X.p[, , 1, ]
X.p.0 <- abind(X.p.0, X.p.0[, , 2, drop = FALSE] * X.p.0[, , 3, drop = FALSE])
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Site covariate on detection ---------------------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4, 0.4)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list()
alpha <- c(-1, 0.5)
p.det <- length(alpha)
p.RE <- list()
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
# X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
# X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
# X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
# X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
trend = X[, , 2])
det.covs <- list(int = X.p[, , , 1],
det.cov.1 = X.p[, , , 2],
trend = X[, , 2])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ trend
det.formula <- ~ trend
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, FALSE)
expect_equal(out$psiRE, FALSE)
})
# Check RE error ----------------------
# test_that("random effect gives error when non-numeric", {
# data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
# data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
# data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# })
# Check RE levels ---------------------
# test_that("random effect levels are correct", {
# expect_equal(sort(unique(unlist(out$p.re.level.names))),
# sort(unique(c(X.p.re))))
# expect_equal(sort(unique(unlist(out$re.level.names))),
# sort(unique(c(X.re))))
# })
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
ar1 = TRUE,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
tmp.data <- data.list
tmp.data$occ.covs$trend[3, ] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- 1
tmp.data$det.covs$trend[1, 1] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- X.0
# X.0.full <- abind::abind(X.0, X.re.0, along = 3)
# dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
X.p.0 <- dat$X.p[, , 1, ]
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Random intercept on occurrence ------------------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list(levels = c(10),
sigma.sq.psi = c(0.8))
alpha <- c(-1)
p.det <- length(alpha)
p.RE <- list()
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
# X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
# X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
occ.factor.1 = X.re[, , 1])
det.covs <- list(int = X.p[, , , 1])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ (1 | occ.factor.1)
det.formula <- ~ 1
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, FALSE)
expect_equal(out$psiRE, TRUE)
})
# Check RE error ----------------------
test_that("random effect gives error when non-numeric", {
data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
#data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
ar1 = TRUE,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
# data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
})
# Check RE levels ---------------------
test_that("random effect levels are correct", {
# expect_equal(sort(unique(unlist(out$p.re.level.names))),
# sort(unique(c(X.p.re))))
expect_equal(sort(unique(unlist(out$re.level.names))),
sort(unique(c(X.re))))
})
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
# tmp.data <- data.list
# tmp.data$occ.covs$trend[3, ] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
# tmp.data <- data.list
# tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- abind::abind(X.0, X.re.0, along = 3)
X.0.full <- abind::abind(X.0, X.re.0, along = 3)
dimnames(X.0.full)[[3]] <- c('(Intercept)', 'occ.factor.1')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
# X.p.0 <- abind::abind(dat$X.p[, , 1, ], dat$X.p.re[, , 1, ], along = 3)
X.p.0 <- array(dat$X.p[, , 1, ], dim = c(J, n.time.max, 1))
# dimnames(X.p.0)[[3]] <- c('(Intercept)', 'det.cov.1', 'det.factor.1', 'det.factor.2')
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Multiple random intercepts on occurrence --------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list(levels = c(10, 15),
sigma.sq.psi = c(0.8, 0.3))
alpha <- c(-1)
p.det <- length(alpha)
p.RE <- list()
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
# X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
# X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
occ.factor.1 = X.re[, , 1],
occ.factor.2 = X.re[, , 2])
det.covs <- list(int = X.p[, , , 1])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ (1 | occ.factor.1) + (1 | occ.factor.2)
det.formula <- ~ 1
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, FALSE)
expect_equal(out$psiRE, TRUE)
})
# Check RE error ----------------------
test_that("random effect gives error when non-numeric", {
data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
#data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
# data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
ar1 = TRUE,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
})
# Check RE levels ---------------------
test_that("random effect levels are correct", {
# expect_equal(sort(unique(unlist(out$p.re.level.names))),
# sort(unique(c(X.p.re))))
expect_equal(sort(unique(unlist(out$re.level.names))),
sort(unique(c(X.re))))
})
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
# tmp.data <- data.list
# tmp.data$occ.covs$trend[3, ] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
# tmp.data <- data.list
# tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- abind::abind(X.0, X.re.0, along = 3)
X.0.full <- abind::abind(X.0, X.re.0, along = 3)
dimnames(X.0.full)[[3]] <- c('(Intercept)', 'occ.factor.1', 'occ.factor.2')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
# X.p.0 <- abind::abind(dat$X.p[, , 1, ], dat$X.p.re[, , 1, ], along = 3)
X.p.0 <- array(dat$X.p[, , 1, ], dim = c(J, n.time.max, 1))
# dimnames(X.p.0)[[3]] <- c('(Intercept)', 'det.cov.1', 'det.factor.1', 'det.factor.2')
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Occurrence REs + covariates ---------------------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4, 0.5)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list(levels = c(10, 15),
sigma.sq.psi = c(0.8, 0.3))
alpha <- c(-1)
p.det <- length(alpha)
p.RE <- list()
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
# X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
# X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
trend = X[, , 2],
occ.factor.1 = X.re[, , 1],
occ.factor.2 = X.re[, , 2])
det.covs <- list(int = X.p[, , , 1])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ trend + (1 | occ.factor.1) + (1 | occ.factor.2)
det.formula <- ~ 1
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, FALSE)
expect_equal(out$psiRE, TRUE)
})
# Check RE error ----------------------
test_that("random effect gives error when non-numeric", {
data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
#data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
# data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
})
# Check RE levels ---------------------
test_that("random effect levels are correct", {
# expect_equal(sort(unique(unlist(out$p.re.level.names))),
# sort(unique(c(X.p.re))))
expect_equal(sort(unique(unlist(out$re.level.names))),
sort(unique(c(X.re))))
})
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
ar1 = TRUE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
tmp.data <- data.list
tmp.data$occ.covs$trend[3, ] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
# tmp.data <- data.list
# tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- abind::abind(X.0, X.re.0, along = 3)
dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend', 'occ.factor.1', 'occ.factor.2')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
# X.p.0 <- abind::abind(dat$X.p[, , 1, ], dat$X.p.re[, , 1, ], along = 3)
X.p.0 <- array(dat$X.p[, , 1, ], dim = c(J, n.time.max, 1))
# dimnames(X.p.0)[[3]] <- c('(Intercept)', 'det.cov.1', 'det.factor.1', 'det.factor.2')
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Occurrence REs + covariates on both -------------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4, 0.5)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list(levels = c(10, 15),
sigma.sq.psi = c(0.8, 0.3))
alpha <- c(-1, 0.5)
p.det <- length(alpha)
p.RE <- list()
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
# X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
# X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
trend = X[, , 2],
occ.factor.1 = X.re[, , 1],
occ.factor.2 = X.re[, , 2])
det.covs <- list(int = X.p[, , , 1],
det.cov.1 = X.p[, , , 2])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ trend + (1 | occ.factor.1) + (1 | occ.factor.2)
det.formula <- ~ det.cov.1
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, FALSE)
expect_equal(out$psiRE, TRUE)
})
# Check RE error ----------------------
test_that("random effect gives error when non-numeric", {
data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
#data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
# data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
})
# Check RE levels ---------------------
test_that("random effect levels are correct", {
# expect_equal(sort(unique(unlist(out$p.re.level.names))),
# sort(unique(c(X.p.re))))
expect_equal(sort(unique(unlist(out$re.level.names))),
sort(unique(c(X.re))))
})
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
tmp.data <- data.list
tmp.data$occ.covs$trend[3, ] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- 1
tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
ar1 = TRUE,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- abind::abind(X.0, X.re.0, along = 3)
dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend', 'occ.factor.1', 'occ.factor.2')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
# X.p.0 <- abind::abind(dat$X.p[, , 1, ], dat$X.p.re[, , 1, ], along = 3)
X.p.0 <- dat$X.p[, , 1, ]
# dimnames(X.p.0)[[3]] <- c('(Intercept)', 'det.cov.1', 'det.factor.1', 'det.factor.2')
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Random intercept on detection -------------------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list()
alpha <- c(-1)
p.det <- length(alpha)
p.RE <- list(levels = c(20),
sigma.sq.p = c(1))
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
# X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
# X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1])
det.covs <- list(int = X.p[, , , 1],
det.factor.1 = X.p.re[, , , 1])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ 1
det.formula <- ~ (1 | det.factor.1)
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, TRUE)
expect_equal(out$psiRE, FALSE)
})
# Check RE error ----------------------
test_that("random effect gives error when non-numeric", {
# data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
# data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
})
# Check RE levels ---------------------
test_that("random effect levels are correct", {
expect_equal(sort(unique(unlist(out$p.re.level.names))),
sort(unique(c(X.p.re))))
# expect_equal(sort(unique(unlist(out$re.level.names))),
# sort(unique(c(X.re))))
})
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
# tmp.data <- data.list
# tmp.data$occ.covs$trend[3, ] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
# tmp.data <- data.list
# tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
# X.0.full <- abind::abind(X.0, X.re.0, along = 3)
# dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend', 'occ.factor.1')
X.0.full <- X.0
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
X.p.0 <- abind::abind(array(dat$X.p[, , 1, ], dim = c(J, n.time.max, 1)),
dat$X.p.re[, , 1, ], along = 3)
dimnames(X.p.0)[[3]] <- c('(Intercept)', 'det.factor.1')
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Multiple random intercepts on detection ---------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list()
alpha <- c(-1)
p.det <- length(alpha)
p.RE <- list(levels = c(20, 15),
sigma.sq.p = c(1, 0.5))
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
# X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
# X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1])
det.covs <- list(int = X.p[, , , 1],
det.factor.1 = X.p.re[, , , 1],
det.factor.2 = X.p.re[, , , 2])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ 1
det.formula <- ~ (1 | det.factor.1) + (1 | det.factor.2)
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
ar1 = TRUE,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, TRUE)
expect_equal(out$psiRE, FALSE)
})
# Check RE error ----------------------
test_that("random effect gives error when non-numeric", {
# data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
# data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
})
# Check RE levels ---------------------
test_that("random effect levels are correct", {
expect_equal(sort(unique(unlist(out$p.re.level.names))),
sort(unique(c(X.p.re))))
# expect_equal(sort(unique(unlist(out$re.level.names))),
# sort(unique(c(X.re))))
})
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
# tmp.data <- data.list
# tmp.data$occ.covs$trend[3, ] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
# tmp.data <- data.list
# tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
# X.0.full <- abind::abind(X.0, X.re.0, along = 3)
# dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend', 'occ.factor.1')
X.0.full <- X.0
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
X.p.0 <- abind::abind(array(dat$X.p[, , 1, ], dim = c(J, n.time.max, 1)),
dat$X.p.re[, , 1, ], along = 3)
dimnames(X.p.0)[[3]] <- c('(Intercept)', 'det.factor.1', 'det.factor.2')
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Random detection intercepts + detection covariates ----------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list()
alpha <- c(-1, 0.5)
p.det <- length(alpha)
p.RE <- list(levels = c(20, 15),
sigma.sq.p = c(1, 0.5))
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
# X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
# X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1])
det.covs <- list(int = X.p[, , , 1],
det.cov.1 = X.p[, , , 2],
det.factor.1 = X.p.re[, , , 1],
det.factor.2 = X.p.re[, , , 2])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ 1
det.formula <- ~ det.cov.1 + (1 | det.factor.1) + (1 | det.factor.2)
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, TRUE)
expect_equal(out$psiRE, FALSE)
})
# Check RE error ----------------------
test_that("random effect gives error when non-numeric", {
# data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
# data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
ar1 = TRUE,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
})
# Check RE levels ---------------------
test_that("random effect levels are correct", {
expect_equal(sort(unique(unlist(out$p.re.level.names))),
sort(unique(c(X.p.re))))
# expect_equal(sort(unique(unlist(out$re.level.names))),
# sort(unique(c(X.re))))
})
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
# tmp.data <- data.list
# tmp.data$occ.covs$trend[3, ] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- 1
tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
# X.0.full <- abind::abind(X.0, X.re.0, along = 3)
# dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend', 'occ.factor.1')
X.0.full <- X.0
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
X.p.0 <- abind::abind(dat$X.p[, , 1, ], dat$X.p.re[, , 1, ], along = 3)
dimnames(X.p.0)[[3]] <- c('(Intercept)', 'det.cov.1', 'det.factor.1', 'det.factor.2')
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Random detection intercepts + covariates --------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4, 0.5)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list()
alpha <- c(-1, 0.5)
p.det <- length(alpha)
p.RE <- list(levels = c(20, 15),
sigma.sq.p = c(1, 0.5))
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
# X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
# X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
trend = X[, , 2])
det.covs <- list(int = X.p[, , , 1],
det.cov.1 = X.p[, , , 2],
det.factor.1 = X.p.re[, , , 1],
det.factor.2 = X.p.re[, , , 2])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ trend
det.formula <- ~ det.cov.1 + (1 | det.factor.1) + (1 | det.factor.2)
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
ar1 = TRUE,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, TRUE)
expect_equal(out$psiRE, FALSE)
})
# Check RE error ----------------------
test_that("random effect gives error when non-numeric", {
# data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
# data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
})
# Check RE levels ---------------------
test_that("random effect levels are correct", {
expect_equal(sort(unique(unlist(out$p.re.level.names))),
sort(unique(c(X.p.re))))
# expect_equal(sort(unique(unlist(out$re.level.names))),
# sort(unique(c(X.re))))
})
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
tmp.data <- data.list
tmp.data$occ.covs$trend[3, ] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- 1
tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
# X.0.full <- abind::abind(X.0, X.re.0, along = 3)
# dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend', 'occ.factor.1')
X.0.full <- X.0
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
X.p.0 <- abind::abind(dat$X.p[, , 1, ], dat$X.p.re[, , 1, ], along = 3)
dimnames(X.p.0)[[3]] <- c('(Intercept)', 'det.cov.1', 'det.factor.1', 'det.factor.2')
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Random intercepts on both -----------------------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list(levels = c(10),
sigma.sq.psi = c(0.8))
alpha <- c(-1)
p.det <- length(alpha)
p.RE <- list(levels = c(20),
sigma.sq.p = c(0.4))
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
occ.factor.1 = X.re[, , 1])
det.covs <- list(int = X.p[, , , 1],
det.factor.1 = X.p.re[, , , 1])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ (1 | occ.factor.1)
det.formula <- ~ (1 | det.factor.1)
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, TRUE)
expect_equal(out$psiRE, TRUE)
})
# Check RE error ----------------------
test_that("random effect gives error when non-numeric", {
data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
ar1 = TRUE,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
})
# Check RE levels ---------------------
test_that("random effect levels are correct", {
expect_equal(sort(unique(unlist(out$p.re.level.names))),
sort(unique(c(X.p.re))))
expect_equal(sort(unique(unlist(out$re.level.names))),
sort(unique(c(X.re))))
})
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
# tmp.data <- data.list
# tmp.data$occ.covs$trend[3, ] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
# tmp.data <- data.list
# tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- abind::abind(X.0, X.re.0, along = 3)
dimnames(X.0.full)[[3]] <- c('(Intercept)', 'occ.factor.1')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
# X.p.0 <- abind::abind(dat$X.p[, , 1, ], dat$X.p.re[, , 1, ], along = 3)
X.p.0 <- abind::abind(array(dat$X.p[, , 1, ], dim = c(J, n.time.max, 1)),
array(dat$X.p.re[, , 1, ], dim = c(J, n.time.max, 1)), along = 3)
dimnames(X.p.0)[[3]] <- c('(Intercept)', 'det.factor.1')
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Random intercepts on both with covariates -------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4, 0.8)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list(levels = c(10),
sigma.sq.psi = c(0.8))
alpha <- c(-1, 0.5)
p.det <- length(alpha)
p.RE <- list(levels = c(20, 15),
sigma.sq.p = c(1, 0.5))
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
trend = X[, , 2],
occ.factor.1 = X.re[, , 1])
det.covs <- list(int = X.p[, , , 1],
det.cov.1 = X.p[, , , 2],
det.factor.1 = X.p.re[, , , 1],
det.factor.2 = X.p.re[, , , 2])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ trend + (1 | occ.factor.1)
det.formula <- ~ det.cov.1 + (1 | det.factor.1) + (1 | det.factor.2)
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, TRUE)
expect_equal(out$psiRE, TRUE)
})
# Check RE error ----------------------
test_that("random effect gives error when non-numeric", {
data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
ar1 = TRUE,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
})
# Check RE levels ---------------------
test_that("random effect levels are correct", {
expect_equal(sort(unique(unlist(out$p.re.level.names))),
sort(unique(c(X.p.re))))
expect_equal(sort(unique(unlist(out$re.level.names))),
sort(unique(c(X.re))))
})
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
tmp.data <- data.list
tmp.data$occ.covs$trend[3, ] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
ar1 = TRUE,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- 1
tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- abind::abind(X.0, X.re.0, along = 3)
dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend', 'occ.factor.1')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
X.p.0 <- abind::abind(dat$X.p[, , 1, ], dat$X.p.re[, , 1, ], along = 3)
dimnames(X.p.0)[[3]] <- c('(Intercept)', 'det.cov.1', 'det.factor.1', 'det.factor.2')
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Third dimension of y != max(n.rep) --------------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
n.rep.max <- 7
beta <- c(0.4, 0.4)
p.occ <- length(beta)
trend <- FALSE
sp.only <- 0
psi.RE <- list()
alpha <- c(-1, 0.5, -0.4)
p.det <- length(alpha)
p.RE <- list()
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu, n.rep.max = n.rep.max)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
# X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
# X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
# X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
# X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
occ.cov.1 = X[, , 2])
det.covs <- list(int = X.p[, , , 1],
det.cov.1 = X.p[, , , 2],
det.cov.2 = X.p[, , , 3])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ occ.cov.1
det.formula <- ~ det.cov.1 + det.cov.2
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, FALSE)
expect_equal(out$psiRE, FALSE)
})
# Check RE error ----------------------
# test_that("random effect gives error when non-numeric", {
# data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
# data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
# data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# })
# Check RE levels ---------------------
# test_that("random effect levels are correct", {
# expect_equal(sort(unique(unlist(out$p.re.level.names))),
# sort(unique(c(X.p.re))))
# expect_equal(sort(unique(unlist(out$re.level.names))),
# sort(unique(c(X.re))))
# })
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
tmp.data <- data.list
tmp.data$occ.covs$occ.cov.1[3, ] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- 1
tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- X.0
# X.0.full <- abind::abind(X.0, X.re.0, along = 3)
# dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
X.p.0 <- dat$X.p[, , 1, ]
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, n.rep.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, n.rep.max))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, n.rep.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, n.rep.max))
})
doserjef/spOccupancy documentation built on Feb. 28, 2025, 1:19 p.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
# Test stPGOcc.R ---------------------------------------------------------
# NNGP --------------------------------------------------------------------
skip_on_cran()
# Intercept only ----------------------------------------------------------
set.seed(150)
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list()
alpha <- c(-1)
p.det <- length(alpha)
p.RE <- list()
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
# X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
# X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
# X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
# X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1])
det.covs <- list(int = X.p[, , , 1])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ 1
det.formula <- ~ 1
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
ar1 = TRUE,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check non-integer n.post -------------
test_that("non-integer n.post", {
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.thin = 13,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
n.omp.threads = 1,
verbose = FALSE))
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, FALSE)
expect_equal(out$psiRE, FALSE)
})
# Check RE error ----------------------
# test_that("random effect gives error when non-numeric", {
# data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
# data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
# data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# })
# Check RE levels ---------------------
# test_that("random effect levels are correct", {
# expect_equal(sort(unique(unlist(out$p.re.level.names))),
# sort(unique(c(X.p.re))))
# expect_equal(sort(unique(unlist(out$re.level.names))),
# sort(unique(c(X.re))))
# })
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
# tmp.data <- data.list
# tmp.data$occ.covs$trend[3, ] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
# tmp.data <- data.list
# tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
# X.0.full <- abind::abind(X.0, X.re.0, along = 3)
# dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend', 'occ.factor.1')
X.0.full <- X.0
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
# X.p.0 <- abind::abind(dat$X.p[, , 1, ], dat$X.p.re[, , 1, ], along = 3)
# dimnames(X.p.0)[[3]] <- c('(Intercept)', 'det.cov.1', 'det.factor.1', 'det.factor.2')
X.p.0 <- array(dat$X.p[, , 1, ], dim = c(J, n.time.max, p.det))
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Occurrence covariate only -----------------------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4, 0.8)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list()
alpha <- c(-1)
p.det <- length(alpha)
p.RE <- list()
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
# X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
# X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
# X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
# X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
trend = X[, , 2])
det.covs <- list(int = X.p[, , , 1])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ trend
det.formula <- ~ 1
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 1,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, FALSE)
expect_equal(out$psiRE, FALSE)
})
# Check RE error ----------------------
# test_that("random effect gives error when non-numeric", {
# data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
# data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
# data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# })
# Check RE levels ---------------------
# test_that("random effect levels are correct", {
# expect_equal(sort(unique(unlist(out$p.re.level.names))),
# sort(unique(c(X.p.re))))
# expect_equal(sort(unique(unlist(out$re.level.names))),
# sort(unique(c(X.re))))
# })
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
ar1 = TRUE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
tmp.data <- data.list
tmp.data$occ.covs$trend[3, ] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
# tmp.data <- data.list
# tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- X.0
# X.0.full <- abind::abind(X.0, X.re.0, along = 3)
dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
X.p.0 <- array(dat$X.p[, , 1, ], dim = c(J, n.time.max, 1))
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Detection covariate only ------------------------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list()
alpha <- c(-1, 0.5, -0.4)
p.det <- length(alpha)
p.RE <- list()
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
# X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
# X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
# X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
# X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1])
det.covs <- list(int = X.p[, , , 1],
det.cov.1 = X.p[, , , 2],
det.cov.2 = X.p[, , , 3])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ 1
det.formula <- ~ det.cov.1 + det.cov.2
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, FALSE)
expect_equal(out$psiRE, FALSE)
})
# Check RE error ----------------------
# test_that("random effect gives error when non-numeric", {
# data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
# data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
# data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# })
# Check RE levels ---------------------
# test_that("random effect levels are correct", {
# expect_equal(sort(unique(unlist(out$p.re.level.names))),
# sort(unique(c(X.p.re))))
# expect_equal(sort(unique(unlist(out$re.level.names))),
# sort(unique(c(X.re))))
# })
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
# tmp.data <- data.list
# tmp.data$occ.covs$trend[3, ] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- 1
tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- X.0
# X.0.full <- abind::abind(X.0, X.re.0, along = 3)
# dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
X.p.0 <- dat$X.p[, , 1, ]
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Covariates on both ------------------------------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4, 0.4)
p.occ <- length(beta)
trend <- FALSE
sp.only <- 0
psi.RE <- list()
alpha <- c(-1, 0.5, -0.4)
p.det <- length(alpha)
p.RE <- list()
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
# X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
# X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
# X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
# X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
occ.cov.1 = X[, , 2])
det.covs <- list(int = X.p[, , , 1],
det.cov.1 = X.p[, , , 2],
det.cov.2 = X.p[, , , 3])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ occ.cov.1
det.formula <- ~ det.cov.1 + det.cov.2
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, FALSE)
expect_equal(out$psiRE, FALSE)
})
# Check RE error ----------------------
# test_that("random effect gives error when non-numeric", {
# data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
# data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
# data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# })
# Check RE levels ---------------------
# test_that("random effect levels are correct", {
# expect_equal(sort(unique(unlist(out$p.re.level.names))),
# sort(unique(c(X.p.re))))
# expect_equal(sort(unique(unlist(out$re.level.names))),
# sort(unique(c(X.re))))
# })
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
tmp.data <- data.list
tmp.data$occ.covs$occ.cov.1[3, ] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- 1
tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- X.0
# X.0.full <- abind::abind(X.0, X.re.0, along = 3)
# dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
X.p.0 <- dat$X.p[, , 1, ]
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Interactions on both ----------------------------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4, 0.4, 0.2)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list()
alpha <- c(-1, 0.5, -0.4)
p.det <- length(alpha)
p.RE <- list()
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
# X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
# X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
# X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
# X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
trend = X[, , 2],
occ.cov.1 = X[, , 3])
det.covs <- list(int = X.p[, , , 1],
det.cov.1 = X.p[, , , 2],
det.cov.2 = X.p[, , , 3])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ trend * occ.cov.1
det.formula <- ~ det.cov.1 * det.cov.2
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
ar1 = TRUE,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, FALSE)
expect_equal(out$psiRE, FALSE)
})
# Check RE error ----------------------
# test_that("random effect gives error when non-numeric", {
# data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
# data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
# data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# })
# Check RE levels ---------------------
# test_that("random effect levels are correct", {
# expect_equal(sort(unique(unlist(out$p.re.level.names))),
# sort(unique(c(X.p.re))))
# expect_equal(sort(unique(unlist(out$re.level.names))),
# sort(unique(c(X.re))))
# })
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
tmp.data <- data.list
tmp.data$occ.covs$trend[3, ] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- 1
tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- abind(X.0, X.0[, , 3, drop = FALSE] * X.0[, , 2, drop = FALSE])
# X.0.full <- abind::abind(X.0, X.re.0, along = 3)
# dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
X.p.0 <- dat$X.p[, , 1, ]
X.p.0 <- abind(X.p.0, X.p.0[, , 2, drop = FALSE] * X.p.0[, , 3, drop = FALSE])
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Site covariate on detection ---------------------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4, 0.4)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list()
alpha <- c(-1, 0.5)
p.det <- length(alpha)
p.RE <- list()
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
# X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
# X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
# X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
# X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
trend = X[, , 2])
det.covs <- list(int = X.p[, , , 1],
det.cov.1 = X.p[, , , 2],
trend = X[, , 2])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ trend
det.formula <- ~ trend
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, FALSE)
expect_equal(out$psiRE, FALSE)
})
# Check RE error ----------------------
# test_that("random effect gives error when non-numeric", {
# data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
# data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
# data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# })
# Check RE levels ---------------------
# test_that("random effect levels are correct", {
# expect_equal(sort(unique(unlist(out$p.re.level.names))),
# sort(unique(c(X.p.re))))
# expect_equal(sort(unique(unlist(out$re.level.names))),
# sort(unique(c(X.re))))
# })
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
ar1 = TRUE,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
tmp.data <- data.list
tmp.data$occ.covs$trend[3, ] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- 1
tmp.data$det.covs$trend[1, 1] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- X.0
# X.0.full <- abind::abind(X.0, X.re.0, along = 3)
# dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
X.p.0 <- dat$X.p[, , 1, ]
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Random intercept on occurrence ------------------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list(levels = c(10),
sigma.sq.psi = c(0.8))
alpha <- c(-1)
p.det <- length(alpha)
p.RE <- list()
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
# X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
# X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
occ.factor.1 = X.re[, , 1])
det.covs <- list(int = X.p[, , , 1])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ (1 | occ.factor.1)
det.formula <- ~ 1
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, FALSE)
expect_equal(out$psiRE, TRUE)
})
# Check RE error ----------------------
test_that("random effect gives error when non-numeric", {
data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
#data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
ar1 = TRUE,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
# data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
})
# Check RE levels ---------------------
test_that("random effect levels are correct", {
# expect_equal(sort(unique(unlist(out$p.re.level.names))),
# sort(unique(c(X.p.re))))
expect_equal(sort(unique(unlist(out$re.level.names))),
sort(unique(c(X.re))))
})
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
# tmp.data <- data.list
# tmp.data$occ.covs$trend[3, ] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
# tmp.data <- data.list
# tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- abind::abind(X.0, X.re.0, along = 3)
X.0.full <- abind::abind(X.0, X.re.0, along = 3)
dimnames(X.0.full)[[3]] <- c('(Intercept)', 'occ.factor.1')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
# X.p.0 <- abind::abind(dat$X.p[, , 1, ], dat$X.p.re[, , 1, ], along = 3)
X.p.0 <- array(dat$X.p[, , 1, ], dim = c(J, n.time.max, 1))
# dimnames(X.p.0)[[3]] <- c('(Intercept)', 'det.cov.1', 'det.factor.1', 'det.factor.2')
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Multiple random intercepts on occurrence --------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list(levels = c(10, 15),
sigma.sq.psi = c(0.8, 0.3))
alpha <- c(-1)
p.det <- length(alpha)
p.RE <- list()
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
# X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
# X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
occ.factor.1 = X.re[, , 1],
occ.factor.2 = X.re[, , 2])
det.covs <- list(int = X.p[, , , 1])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ (1 | occ.factor.1) + (1 | occ.factor.2)
det.formula <- ~ 1
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, FALSE)
expect_equal(out$psiRE, TRUE)
})
# Check RE error ----------------------
test_that("random effect gives error when non-numeric", {
data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
#data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
# data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
ar1 = TRUE,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
})
# Check RE levels ---------------------
test_that("random effect levels are correct", {
# expect_equal(sort(unique(unlist(out$p.re.level.names))),
# sort(unique(c(X.p.re))))
expect_equal(sort(unique(unlist(out$re.level.names))),
sort(unique(c(X.re))))
})
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
# tmp.data <- data.list
# tmp.data$occ.covs$trend[3, ] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
# tmp.data <- data.list
# tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- abind::abind(X.0, X.re.0, along = 3)
X.0.full <- abind::abind(X.0, X.re.0, along = 3)
dimnames(X.0.full)[[3]] <- c('(Intercept)', 'occ.factor.1', 'occ.factor.2')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
# X.p.0 <- abind::abind(dat$X.p[, , 1, ], dat$X.p.re[, , 1, ], along = 3)
X.p.0 <- array(dat$X.p[, , 1, ], dim = c(J, n.time.max, 1))
# dimnames(X.p.0)[[3]] <- c('(Intercept)', 'det.cov.1', 'det.factor.1', 'det.factor.2')
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Occurrence REs + covariates ---------------------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4, 0.5)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list(levels = c(10, 15),
sigma.sq.psi = c(0.8, 0.3))
alpha <- c(-1)
p.det <- length(alpha)
p.RE <- list()
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
# X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
# X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
trend = X[, , 2],
occ.factor.1 = X.re[, , 1],
occ.factor.2 = X.re[, , 2])
det.covs <- list(int = X.p[, , , 1])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ trend + (1 | occ.factor.1) + (1 | occ.factor.2)
det.formula <- ~ 1
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, FALSE)
expect_equal(out$psiRE, TRUE)
})
# Check RE error ----------------------
test_that("random effect gives error when non-numeric", {
data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
#data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
# data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
})
# Check RE levels ---------------------
test_that("random effect levels are correct", {
# expect_equal(sort(unique(unlist(out$p.re.level.names))),
# sort(unique(c(X.p.re))))
expect_equal(sort(unique(unlist(out$re.level.names))),
sort(unique(c(X.re))))
})
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
ar1 = TRUE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
tmp.data <- data.list
tmp.data$occ.covs$trend[3, ] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
# tmp.data <- data.list
# tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- abind::abind(X.0, X.re.0, along = 3)
dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend', 'occ.factor.1', 'occ.factor.2')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
# X.p.0 <- abind::abind(dat$X.p[, , 1, ], dat$X.p.re[, , 1, ], along = 3)
X.p.0 <- array(dat$X.p[, , 1, ], dim = c(J, n.time.max, 1))
# dimnames(X.p.0)[[3]] <- c('(Intercept)', 'det.cov.1', 'det.factor.1', 'det.factor.2')
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Occurrence REs + covariates on both -------------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4, 0.5)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list(levels = c(10, 15),
sigma.sq.psi = c(0.8, 0.3))
alpha <- c(-1, 0.5)
p.det <- length(alpha)
p.RE <- list()
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
# X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
# X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
trend = X[, , 2],
occ.factor.1 = X.re[, , 1],
occ.factor.2 = X.re[, , 2])
det.covs <- list(int = X.p[, , , 1],
det.cov.1 = X.p[, , , 2])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ trend + (1 | occ.factor.1) + (1 | occ.factor.2)
det.formula <- ~ det.cov.1
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, FALSE)
expect_equal(out$psiRE, TRUE)
})
# Check RE error ----------------------
test_that("random effect gives error when non-numeric", {
data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
#data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
# data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
})
# Check RE levels ---------------------
test_that("random effect levels are correct", {
# expect_equal(sort(unique(unlist(out$p.re.level.names))),
# sort(unique(c(X.p.re))))
expect_equal(sort(unique(unlist(out$re.level.names))),
sort(unique(c(X.re))))
})
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
tmp.data <- data.list
tmp.data$occ.covs$trend[3, ] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- 1
tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
ar1 = TRUE,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- abind::abind(X.0, X.re.0, along = 3)
dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend', 'occ.factor.1', 'occ.factor.2')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
# X.p.0 <- abind::abind(dat$X.p[, , 1, ], dat$X.p.re[, , 1, ], along = 3)
X.p.0 <- dat$X.p[, , 1, ]
# dimnames(X.p.0)[[3]] <- c('(Intercept)', 'det.cov.1', 'det.factor.1', 'det.factor.2')
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Random intercept on detection -------------------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list()
alpha <- c(-1)
p.det <- length(alpha)
p.RE <- list(levels = c(20),
sigma.sq.p = c(1))
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
# X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
# X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1])
det.covs <- list(int = X.p[, , , 1],
det.factor.1 = X.p.re[, , , 1])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ 1
det.formula <- ~ (1 | det.factor.1)
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, TRUE)
expect_equal(out$psiRE, FALSE)
})
# Check RE error ----------------------
test_that("random effect gives error when non-numeric", {
# data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
# data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
})
# Check RE levels ---------------------
test_that("random effect levels are correct", {
expect_equal(sort(unique(unlist(out$p.re.level.names))),
sort(unique(c(X.p.re))))
# expect_equal(sort(unique(unlist(out$re.level.names))),
# sort(unique(c(X.re))))
})
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
# tmp.data <- data.list
# tmp.data$occ.covs$trend[3, ] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
# tmp.data <- data.list
# tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
# X.0.full <- abind::abind(X.0, X.re.0, along = 3)
# dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend', 'occ.factor.1')
X.0.full <- X.0
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
X.p.0 <- abind::abind(array(dat$X.p[, , 1, ], dim = c(J, n.time.max, 1)),
dat$X.p.re[, , 1, ], along = 3)
dimnames(X.p.0)[[3]] <- c('(Intercept)', 'det.factor.1')
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Multiple random intercepts on detection ---------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list()
alpha <- c(-1)
p.det <- length(alpha)
p.RE <- list(levels = c(20, 15),
sigma.sq.p = c(1, 0.5))
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
# X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
# X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1])
det.covs <- list(int = X.p[, , , 1],
det.factor.1 = X.p.re[, , , 1],
det.factor.2 = X.p.re[, , , 2])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ 1
det.formula <- ~ (1 | det.factor.1) + (1 | det.factor.2)
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
ar1 = TRUE,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, TRUE)
expect_equal(out$psiRE, FALSE)
})
# Check RE error ----------------------
test_that("random effect gives error when non-numeric", {
# data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
# data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
})
# Check RE levels ---------------------
test_that("random effect levels are correct", {
expect_equal(sort(unique(unlist(out$p.re.level.names))),
sort(unique(c(X.p.re))))
# expect_equal(sort(unique(unlist(out$re.level.names))),
# sort(unique(c(X.re))))
})
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
# tmp.data <- data.list
# tmp.data$occ.covs$trend[3, ] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
# tmp.data <- data.list
# tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
# X.0.full <- abind::abind(X.0, X.re.0, along = 3)
# dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend', 'occ.factor.1')
X.0.full <- X.0
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
X.p.0 <- abind::abind(array(dat$X.p[, , 1, ], dim = c(J, n.time.max, 1)),
dat$X.p.re[, , 1, ], along = 3)
dimnames(X.p.0)[[3]] <- c('(Intercept)', 'det.factor.1', 'det.factor.2')
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Random detection intercepts + detection covariates ----------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list()
alpha <- c(-1, 0.5)
p.det <- length(alpha)
p.RE <- list(levels = c(20, 15),
sigma.sq.p = c(1, 0.5))
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
# X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
# X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1])
det.covs <- list(int = X.p[, , , 1],
det.cov.1 = X.p[, , , 2],
det.factor.1 = X.p.re[, , , 1],
det.factor.2 = X.p.re[, , , 2])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ 1
det.formula <- ~ det.cov.1 + (1 | det.factor.1) + (1 | det.factor.2)
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, TRUE)
expect_equal(out$psiRE, FALSE)
})
# Check RE error ----------------------
test_that("random effect gives error when non-numeric", {
# data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
# data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
ar1 = TRUE,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
})
# Check RE levels ---------------------
test_that("random effect levels are correct", {
expect_equal(sort(unique(unlist(out$p.re.level.names))),
sort(unique(c(X.p.re))))
# expect_equal(sort(unique(unlist(out$re.level.names))),
# sort(unique(c(X.re))))
})
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
# tmp.data <- data.list
# tmp.data$occ.covs$trend[3, ] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- 1
tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
# X.0.full <- abind::abind(X.0, X.re.0, along = 3)
# dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend', 'occ.factor.1')
X.0.full <- X.0
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
X.p.0 <- abind::abind(dat$X.p[, , 1, ], dat$X.p.re[, , 1, ], along = 3)
dimnames(X.p.0)[[3]] <- c('(Intercept)', 'det.cov.1', 'det.factor.1', 'det.factor.2')
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Random detection intercepts + covariates --------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4, 0.5)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list()
alpha <- c(-1, 0.5)
p.det <- length(alpha)
p.RE <- list(levels = c(20, 15),
sigma.sq.p = c(1, 0.5))
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
# X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
# X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
trend = X[, , 2])
det.covs <- list(int = X.p[, , , 1],
det.cov.1 = X.p[, , , 2],
det.factor.1 = X.p.re[, , , 1],
det.factor.2 = X.p.re[, , , 2])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ trend
det.formula <- ~ det.cov.1 + (1 | det.factor.1) + (1 | det.factor.2)
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
ar1 = TRUE,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, TRUE)
expect_equal(out$psiRE, FALSE)
})
# Check RE error ----------------------
test_that("random effect gives error when non-numeric", {
# data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
# data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
})
# Check RE levels ---------------------
test_that("random effect levels are correct", {
expect_equal(sort(unique(unlist(out$p.re.level.names))),
sort(unique(c(X.p.re))))
# expect_equal(sort(unique(unlist(out$re.level.names))),
# sort(unique(c(X.re))))
})
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
tmp.data <- data.list
tmp.data$occ.covs$trend[3, ] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- 1
tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
# X.0.full <- abind::abind(X.0, X.re.0, along = 3)
# dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend', 'occ.factor.1')
X.0.full <- X.0
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
X.p.0 <- abind::abind(dat$X.p[, , 1, ], dat$X.p.re[, , 1, ], along = 3)
dimnames(X.p.0)[[3]] <- c('(Intercept)', 'det.cov.1', 'det.factor.1', 'det.factor.2')
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Random intercepts on both -----------------------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list(levels = c(10),
sigma.sq.psi = c(0.8))
alpha <- c(-1)
p.det <- length(alpha)
p.RE <- list(levels = c(20),
sigma.sq.p = c(0.4))
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
occ.factor.1 = X.re[, , 1])
det.covs <- list(int = X.p[, , , 1],
det.factor.1 = X.p.re[, , , 1])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ (1 | occ.factor.1)
det.formula <- ~ (1 | det.factor.1)
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, TRUE)
expect_equal(out$psiRE, TRUE)
})
# Check RE error ----------------------
test_that("random effect gives error when non-numeric", {
data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
ar1 = TRUE,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
})
# Check RE levels ---------------------
test_that("random effect levels are correct", {
expect_equal(sort(unique(unlist(out$p.re.level.names))),
sort(unique(c(X.p.re))))
expect_equal(sort(unique(unlist(out$re.level.names))),
sort(unique(c(X.re))))
})
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
# tmp.data <- data.list
# tmp.data$occ.covs$trend[3, ] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
# tmp.data <- data.list
# tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
# expect_error(stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = tmp.data,
# n.batch = 40,
# batch.length = batch.length,
# cov.model = "exponential",
# tuning = tuning.list,
# NNGP = TRUE,
# verbose = FALSE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- abind::abind(X.0, X.re.0, along = 3)
dimnames(X.0.full)[[3]] <- c('(Intercept)', 'occ.factor.1')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
# X.p.0 <- abind::abind(dat$X.p[, , 1, ], dat$X.p.re[, , 1, ], along = 3)
X.p.0 <- abind::abind(array(dat$X.p[, , 1, ], dim = c(J, n.time.max, 1)),
array(dat$X.p.re[, , 1, ], dim = c(J, n.time.max, 1)), along = 3)
dimnames(X.p.0)[[3]] <- c('(Intercept)', 'det.factor.1')
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Random intercepts on both with covariates -------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
beta <- c(0.4, 0.8)
p.occ <- length(beta)
trend <- TRUE
sp.only <- 0
psi.RE <- list(levels = c(10),
sigma.sq.psi = c(0.8))
alpha <- c(-1, 0.5)
p.det <- length(alpha)
p.RE <- list(levels = c(20, 15),
sigma.sq.p = c(1, 0.5))
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
trend = X[, , 2],
occ.factor.1 = X.re[, , 1])
det.covs <- list(int = X.p[, , , 1],
det.cov.1 = X.p[, , , 2],
det.factor.1 = X.p.re[, , , 1],
det.factor.2 = X.p.re[, , , 2])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ trend + (1 | occ.factor.1)
det.formula <- ~ det.cov.1 + (1 | det.factor.1) + (1 | det.factor.2)
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, TRUE)
expect_equal(out$psiRE, TRUE)
})
# Check RE error ----------------------
test_that("random effect gives error when non-numeric", {
data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
ar1 = TRUE,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
expect_error(out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
accept.rate = 0.43,
priors = prior.list,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.thin = 2,
n.chains = 1))
})
# Check RE levels ---------------------
test_that("random effect levels are correct", {
expect_equal(sort(unique(unlist(out$p.re.level.names))),
sort(unique(c(X.p.re))))
expect_equal(sort(unique(unlist(out$re.level.names))),
sort(unique(c(X.re))))
})
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
tmp.data <- data.list
tmp.data$occ.covs$trend[3, ] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
ar1 = TRUE,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- 1
tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- abind::abind(X.0, X.re.0, along = 3)
dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend', 'occ.factor.1')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
X.p.0 <- abind::abind(dat$X.p[, , 1, ], dat$X.p.re[, , 1, ], along = 3)
dimnames(X.p.0)[[3]] <- c('(Intercept)', 'det.cov.1', 'det.factor.1', 'det.factor.2')
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, max(n.rep, na.rm = TRUE)))
})
# Third dimension of y != max(n.rep) --------------------------------------
# Sites
J.x <- 10
J.y <- 10
J <- J.x * J.y
n.time <- sample(2:10, J, replace = TRUE)
n.time.max <- max(n.time)
n.rep <- matrix(NA, J, max(n.time))
for (j in 1:J) {
n.rep[j, 1:n.time[j]] <- sample(1:4, n.time[j], replace = TRUE)
}
n.rep.max <- 7
beta <- c(0.4, 0.4)
p.occ <- length(beta)
trend <- FALSE
sp.only <- 0
psi.RE <- list()
alpha <- c(-1, 0.5, -0.4)
p.det <- length(alpha)
p.RE <- list()
phi <- 3 / .7
sigma.sq <- 2
nu <- 2
dat <- simTOcc(J.x = J.x, J.y = J.y, n.time = n.time, n.rep = n.rep,
beta = beta, alpha = alpha, sp.only = sp.only, trend = trend,
psi.RE = psi.RE, p.RE = p.RE, sp = TRUE, sigma.sq = sigma.sq,
phi = phi, cov.model = 'matern', nu = nu, n.rep.max = n.rep.max)
# Subset data for prediction
pred.indx <- sample(1:J, round(J * .25), replace = FALSE)
y <- dat$y[-pred.indx, , , drop = FALSE]
# Occupancy covariates
X <- dat$X[-pred.indx, , , drop = FALSE]
# X.re <- dat$X.re[-pred.indx, , , drop = FALSE]
# Prediction covariates
X.0 <- dat$X[pred.indx, , , drop = FALSE]
# X.re.0 <- dat$X.re[pred.indx, , , drop = FALSE]
# Detection covariates
X.p <- dat$X.p[-pred.indx, , , , drop = FALSE]
# X.p.re <- dat$X.p.re[-pred.indx, , , , drop = FALSE]
X.p.0 <- dat$X.p[pred.indx, , , , drop = FALSE]
# X.p.re.0 <- dat$X.p.re[pred.indx, , , , drop = FALSE]
coords <- as.matrix(dat$coords[-pred.indx, ])
coords.0 <- as.matrix(dat$coords[pred.indx, ])
psi.0 <- dat$psi[pred.indx, ]
w.0 <- dat$w[pred.indx]
occ.covs <- list(int = X[, , 1],
occ.cov.1 = X[, , 2])
det.covs <- list(int = X.p[, , , 1],
det.cov.1 = X.p[, , , 2],
det.cov.2 = X.p[, , , 3])
data.list <- list(y = y, coords = coords, occ.covs = occ.covs, det.covs = det.covs)
# Priors
prior.list <- list(beta.normal = list(mean = 0, var = 2.72),
alpha.normal = list(mean = 0, var = 2.72),
nu.unif = c(0.5, 2.5))
# Starting values
z.init <- apply(y, c(1, 2), function(a) as.numeric(sum(a, na.rm = TRUE) > 0))
inits.list <- list(alpha = 0, beta = 0, z = z.init)
# Tuning
tuning.list <- list(phi = 1, nu = 1, rho = 1)
batch.length <- 25
n.batch <- 10
n.report <- 10
n.samples <- batch.length * n.batch
occ.formula <- ~ occ.cov.1
det.formula <- ~ det.cov.1 + det.cov.2
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
batch.length = batch.length,
n.batch = n.batch,
priors = prior.list,
accept.rate = 0.43,
cov.model = "matern",
tuning = tuning.list,
n.omp.threads = 1,
verbose = FALSE,
NNGP = TRUE,
n.neighbors = 10,
n.report = n.report,
n.burn = 100,
n.chains = 2,
n.thin = 2,
k.fold = 2,
k.fold.threads = 2)
# Test to make sure it worked ---------
test_that("out is of class stPGOcc", {
expect_s3_class(out, "stPGOcc")
})
# Check cross-validation --------------
test_that("cross-validation works", {
expect_equal(length(out$k.fold.deviance), 1)
expect_type(out$k.fold.deviance, "double")
expect_gt(out$k.fold.deviance, 0)
})
# Check random effects ----------------
test_that("random effects are empty", {
expect_equal(out$pRE, FALSE)
expect_equal(out$psiRE, FALSE)
})
# Check RE error ----------------------
# test_that("random effect gives error when non-numeric", {
# data.list$occ.covs$occ.factor.1 <- factor(data.list$occ.covs$occ.factor.1)
# data.list$det.covs$det.factor.1 <- factor(data.list$det.covs$det.factor.1)
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# data.list$occ.covs$occ.factor.1 <- as.character(factor(data.list$occ.covs$occ.factor.1))
# data.list$det.covs$det.factor.1 <- as.character(factor(data.list$det.covs$det.factor.1))
# expect_error(out <- stPGOcc(occ.formula = occ.formula,
# det.formula = det.formula,
# data = data.list,
# inits = inits.list,
# n.batch = n.batch,
# batch.length = batch.length,
# accept.rate = 0.43,
# priors = prior.list,
# cov.model = "matern",
# tuning = tuning.list,
# n.omp.threads = 1,
# verbose = FALSE,
# NNGP = TRUE,
# n.neighbors = 5,
# search.type = 'cb',
# n.report = 10,
# n.burn = 100,
# n.thin = 2,
# n.chains = 1))
# })
# Check RE levels ---------------------
# test_that("random effect levels are correct", {
# expect_equal(sort(unique(unlist(out$p.re.level.names))),
# sort(unique(c(X.p.re))))
# expect_equal(sort(unique(unlist(out$re.level.names))),
# sort(unique(c(X.re))))
# })
# Check output data output is correct -
test_that("out$y == y", {
expect_equal(out$y, y)
})
test_that("default priors and inits work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("all correlation functions work", {
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "gaussian",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
n.batch = 40,
batch.length = batch.length,
cov.model = "spherical",
tuning = list(phi = 0.5),
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
test_that("verbose prints to the screen", {
expect_output(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = data.list,
inits = inits.list,
n.batch = n.batch,
batch.length = batch.length,
priors = prior.list,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
})
# Check summary -----------------------
test_that("summary works", {
expect_output(summary(out))
})
# Check missing values ----------------
test_that("missing value error handling works", {
tmp.data <- data.list
tmp.data$occ.covs$occ.cov.1[3, ] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- 1
tmp.data$det.covs$det.cov.1[1, 1, 1] <- NA
expect_error(stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1))
tmp.data <- data.list
tmp.data$y[1, 1, 1] <- NA
out <- stPGOcc(occ.formula = occ.formula,
det.formula = det.formula,
data = tmp.data,
n.batch = 40,
batch.length = batch.length,
cov.model = "exponential",
tuning = tuning.list,
NNGP = TRUE,
verbose = FALSE,
n.neighbors = 5,
search.type = 'cb',
n.report = 10,
n.burn = 100,
n.chains = 1)
expect_s3_class(out, "stPGOcc")
})
# Check waicOcc -----------------------
test_that("waicOCC works for stPGOcc", {
# as.vector gets rid of names
waic.out <- as.vector(waicOcc(out))
expect_equal(length(waic.out), 3)
expect_equal(waic.out[3], -2 * (waic.out[1] - waic.out[2]))
})
# Check fitted ------------------------
test_that("fitted works for stPGOcc", {
fitted.out <- fitted(out)
expect_equal(length(fitted.out), 2)
})
# Check predictions -------------------
test_that("predict works for stPGOcc", {
X.0.full <- X.0
# X.0.full <- abind::abind(X.0, X.re.0, along = 3)
# dimnames(X.0.full)[[3]] <- c('(Intercept)', 'trend')
pred.out <- predict(out, X.0.full, coords.0, verbose = FALSE, t.cols = 1:n.time.max)
expect_type(pred.out, "list")
expect_equal(dim(pred.out$psi.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
expect_equal(dim(pred.out$z.0.samples), c(out$n.post * out$n.chains, nrow(X.0.full), n.time.max))
})
test_that("detection prediction works", {
X.p.0 <- dat$X.p[, , 1, ]
pred.out <- predict(out, X.p.0, type = 'detection', t.cols = 1:n.time.max)
expect_type(pred.out, 'list')
expect_equal(dim(pred.out$p.0.samples), c(out$n.post * out$n.chains, J, n.time.max))
})
# Check PPCs --------------------------
test_that("posterior predictive checks work for tPGOcc", {
J.fit <- nrow(X)
n.post.samples <- out$n.post * out$n.chains
ppc.out <- ppcOcc(out, 'chi-square', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, n.rep.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, n.rep.max))
ppc.out <- ppcOcc(out, 'chi-square', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 1)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, J.fit, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, J.fit, n.time.max))
ppc.out <- ppcOcc(out, 'freeman-tukey', 2)
expect_type(ppc.out, "list")
expect_equal(dim(ppc.out$fit.y), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.rep), c(n.post.samples, n.time.max))
expect_equal(dim(ppc.out$fit.y.group.quants), c(5, n.time.max, n.rep.max))
expect_equal(dim(ppc.out$fit.y.rep.group.quants), c(5, n.time.max, n.rep.max))
})
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.