tests/testthat/test-lolog-fit.R

In lolog: Latent Order Logistic Graph Models

context("LOLOG Fit")
#library(lolog)
library(testthat)
library(ergm)


test_that("lolog_fit", {
  data(sampson)
  fit <- lolog(
    samplike ~ edges() + triangles(),
    samplike ~ edges() + triangles(),
    theta = c(-1.2304996, 0),
    nsamp = 200,
    includeOrderIndependent = FALSE,
    verbose = FALSE
  )
  expect_true(all(fit$theta > c(-1.28, .18)) &
                all(fit$theta < c(-1.22, .22)))
  
  # Test Dyad Independent
  efit <- ergm(samplike ~ edges + nodematch("group"))
  vfit <- lologVariational(samplike ~ edges() + nodeMatch("group"))
  expect_true(all(round(efit$coef - vfit$theta, 6) == 0))
  
  gmmfit <- lolog(
    samplike ~ edges() + nodeMatch("group"),
    theta = vfit$theta,
    nsamp = 500,
    verbose = FALSE
  )
  svar <- summary(vfit)
  sgmm <- summary(gmmfit)
  expect_true(all(abs(svar$se - sgmm$se) < .04))
  expect_true(all(abs(svar$theta - sgmm$theta) < .02))
  
  
  # order dependent
  data(flo)
  flomarriage <- network(flo, directed = FALSE)
  fit <- lolog(
    flomarriage ~ edges() + preferentialAttachment(),
    flomarriage ~ star(2),
    theta = c(-1.54998018, 0),
    nsamp = 400,
    verbose = FALSE
  )
  expect_true(all(fit$theta > c(-1.9, 0)) &
                all(fit$theta < c(-1.1, .2)))
  
  fit <- lolog(
    flomarriage ~ edges() + preferentialAttachment(),
    flomarriage ~ star(2:3),
    theta = c(-1.54998018, 0),
    nsamp = 400,
    verbose = FALSE
  )
  expect_true(all(dim(fit$grad) == c(3,2)))
  expect_equivalent(summary(fit)$theta, fit$theta)
  plot(fit)
})



test_that("lolog_fit_parallel", {
  library(parallel)
  cluster <- parallel::makeCluster(2)
  # order dependent
  data(flo)
  flomarriage <- network(flo, directed = FALSE)
  fit <- lolog(
    flomarriage ~ edges() + preferentialAttachment(),
    flomarriage ~ star(2L),
    theta = c(-1.54998018, 0),
    nsamp = 400,
    cluster = cluster,
    verbose = FALSE
  )
  expect_true(all(fit$theta > c(-1.9, -.1)) &
                all(fit$theta < c(-1.1, .2)))
  
  
  fit <- lolog(
    flomarriage ~ edges() + star(2),
    theta = c(-1.54998018, 0),
    nsamp = 200,
    cluster = cluster,
    verbose = FALSE
  )
  expect_true(all(fit$theta > c(-1.8, -.1)) &
                all(fit$theta < c(-1.4, .2)))
  parallel::stopCluster(cluster)
})



test_that("lolog_target_stats", {
  data(sampson)
  fit <- lolog(
    samplike ~ edges(),
    theta = c(-1.2304996),
    nsamp = 200,
    targetStats=40,
    verbose = FALSE
  )
  expect_true(round(fit$theta,1) == -1.9)
})