tests/testthat/test_estn.R
In LeeMorinUCF/FCVAR: Estimation and Inference for the Fractionally Cointegrated VAR
context("Estimation")
test_that("estimation options have correct defaults", {
load(file = 'soln_estn/opt.RData')
expect_equal(FCVARoptions(), opt)
})
test_that("estimation options are updated correctly", {
load(file = 'soln_estn/newOpt.RData')
opt <- FCVARoptions(
gridSearch = 0, # Disable grid search in optimization.
dbMin = c(0.01, 0.01), # Set lower bound for d,b.
dbMax = c(2.00, 2.00), # Set upper bound for d,b.
constrained = 0 # Impose restriction dbMax >= d >= b >= dbMin ? 1 <- yes, 0 <- no.
)
expect_equal(FCVARoptionUpdates(opt, p = 3, r = 1), newOpt)
})
test_that("default bounds on parameter space are correct", {
load(file = 'soln_estn/UB_LB_bounds_def.RData')
opt <- FCVARoptions()
expect_equal(GetBounds(opt), UB_LB_bounds)
})
test_that("modified bounds on parameter space are correct", {
load(file = 'soln_estn/UB_LB_bounds_mod.RData')
opt <- FCVARoptions(
dbMin = c(0.01, 0.01), # Set lower bound for d,b.
dbMax = c(2.00, 2.00) # Set upper bound for d,b.
)
expect_equal(GetBounds(opt), UB_LB_bounds)
})
test_that("unrestricted estimation results and output are correct", {
load(file = 'soln_estn/results_m1.RData')
results_text_soln <- readLines('soln_estn/results_m1.txt')
opt <- FCVARoptions(
unc_optim_control = list(maxit = 1000, # Reduce tolerance.
reltol = 1e-10),
con_optim_control = list(maxit = 1000, # Reduce tolerance.
pgtol = 1e-10),
hess_delta = 10^(-3), # Stable for testing on 32-bit, 386 architecture.
gridSearch = 0, # Disable grid search in optimization.
dbMin = c(0.01, 0.01), # Set lower bound for d,b.
dbMax = c(2.00, 2.00), # Set upper bound for d,b.
constrained = 0, # Impose restriction dbMax >= d >= b >= dbMin ? 1 <- yes, 0 <- no.
plotRoots = 0 # Don't create plots for tests.
)
x <- votingJNP2014[, c("lib", "ir_can", "un_can")]
# results_test <- FCVARestn(x, k = 2, r = 1, opt)
capture.output(results_test <- FCVARestn(x, k = 2, r = 1, opt), file = 'soln_estn/temp.txt')
results_text <- readLines('soln_estn/temp.txt')
# Note that the inverse Hessian matrix has cross-platform differences in
# mean relative difference of 1.968147e-08, which is not a problem.
# results$NegInvHessian <- round(results$NegInvHessian, 6)
# results_test$NegInvHessian <- round(results_test$NegInvHessian, 6)
# # Similarly for other estimates.
# results$SE$db <- round(results$SE$db, 6)
# results_test$SE$db <- round(results_test$SE$db, 6)
# results$SE$muHat <- round(results$SE$muHat, 6)
# results_test$SE$muHat <- round(results_test$SE$muHat, 6)
# results$SE$alphaHat <- round(results$SE$alphaHat, 6)
# results_test$SE$alphaHat <- round(results_test$SE$alphaHat, 6)
# # results$SE$gammaHat <- round(results$SE$gammaHat, 6)
# # results_test$SE$gammaHat <- round(results_test$SE$gammaHat, 6)
# Conduct separate tests on Hessian and standard errors.
results_NegInvHessian <- results$NegInvHessian
results$NegInvHessian <- NULL
results_test_NegInvHessian <- results_test$NegInvHessian
results_test$NegInvHessian <- NULL
results_SE <- results$SE
results$SE <- NULL
results_test_SE <- results_test$SE
results_test$SE <- NULL
# Accuracy can only be so high on 32-bit platforms.
expect_equal(results_test, results)
expect_equal(results_test_NegInvHessian, results_NegInvHessian, tolerance = 10^(-5))
expect_equal(results_test_SE, results_SE, tolerance = 10^(-5))
expect_equal(results_text, results_text_soln)
})
test_that("restricted estimation with d = b = 1 is correct", {
load(file = 'soln_estn/results_m1r124.RData')
m1r1_text_soln <- readLines('soln_estn/results_m1r1.txt')
opt <- FCVARoptions(
unc_optim_control = list(maxit = 1000, # Reduce tolerance.
reltol = 1e-10),
con_optim_control = list(maxit = 1000, # Reduce tolerance.
pgtol = 1e-10),
hess_delta = 10^(-3), # Stable for testing on 32-bit, 386 architecture.
gridSearch = 0, # Disable grid search in optimization.
dbMin = c(0.01, 0.01), # Set lower bound for d,b.
dbMax = c(2.00, 2.00), # Set upper bound for d,b.
constrained = 0, # Impose restriction dbMax >= d >= b >= dbMin ? 1 <- yes, 0 <- no.
plotRoots = 0 # Don't create plots for tests.
)
x <- votingJNP2014[, c("lib", "ir_can", "un_can")]
opt1 <- opt
opt1$R_psi <- matrix(c(1, 0), nrow = 1, ncol = 2)
opt1$r_psi <- 1
capture.output(m1r1_test <- FCVARestn(x, k = 2, r = 1, opt1),
file = 'soln_estn/temp.txt')
m1r1_text <- readLines('soln_estn/temp.txt')
# Note that the inverse Hessian matrix has cross-platform differences in
# mean relative difference on the order of 1.0e-08, which is not a problem.
# m1r1$NegInvHessian <- round(m1r1$NegInvHessian, 6)
# m1r1_test$NegInvHessian <- round(m1r1_test$NegInvHessian, 6)
# # Similarly for other estimates.
# m1r1$SE$db <- round(m1r1$SE$db, 6)
# m1r1_test$SE$db <- round(m1r1_test$SE$db, 6)
# m1r1$SE$muHat <- round(m1r1$SE$muHat, 6)
# m1r1_test$SE$muHat <- round(m1r1_test$SE$muHat, 6)
# m1r1$SE$alphaHat <- round(m1r1$SE$alphaHat, 6)
# m1r1_test$SE$alphaHat <- round(m1r1_test$SE$alphaHat, 6)
# # m1r1$SE$gammaHat <- round(m1r1$SE$gammaHat, 6)
# # m1r1_test$SE$gammaHat <- round(m1r1_test$SE$gammaHat, 6)
# Conduct separate tests on Hessian and standard errors.
m1r1_NegInvHessian <- m1r1$NegInvHessian
m1r1$NegInvHessian <- NULL
m1r1_test_NegInvHessian <- m1r1_test$NegInvHessian
m1r1_test$NegInvHessian <- NULL
m1r1_SE <- m1r1$SE
m1r1$SE <- NULL
m1r1_test_SE <- m1r1_test$SE
m1r1_test$SE <- NULL
# Accuracy can only be so high on 32-bit platforms.
expect_equal(m1r1_test, m1r1)
expect_equal(m1r1_test_NegInvHessian, m1r1_NegInvHessian, tolerance = 10^(-5))
expect_equal(m1r1_test_SE, m1r1_SE, tolerance = 10^(-5))
# Exact equality (without tolerance) is not expected on all platforms.
skip_on_cran()
expect_equal(m1r1_text, m1r1_text_soln)
})
test_that("restricted estimation with R_Beta <- c(1, 0, 0) is correct", {
load(file = 'soln_estn/results_m1r124.RData')
m1r2_text_soln <- readLines('soln_estn/results_m1r2.txt')
opt <- FCVARoptions(
unc_optim_control = list(maxit = 1000, # Reduce tolerance.
reltol = 1e-10),
con_optim_control = list(maxit = 1000, # Reduce tolerance.
pgtol = 1e-10),
hess_delta = 10^(-3), # Stable for testing on 32-bit, 386 architecture.
gridSearch = 0, # Disable grid search in optimization.
dbMin = c(0.01, 0.01), # Set lower bound for d,b.
dbMax = c(2.00, 2.00), # Set upper bound for d,b.
constrained = 0, # Impose restriction dbMax >= d >= b >= dbMin ? 1 <- yes, 0 <- no.
plotRoots = 0 # Don't create plots for tests.
)
x <- votingJNP2014[, c("lib", "ir_can", "un_can")]
opt1 <- opt
opt1$R_Beta <- matrix(c(1, 0, 0), nrow = 1, ncol = 3)
# opt1$db0 <- c(0.67, 0.67) # Set starting values for optimization algorithm.
capture.output(m1r2_test <- FCVARestn(x, k = 2, r = 1, opt1),
file = 'soln_estn/temp.txt')
m1r2_text <- readLines('soln_estn/temp.txt')
# Note that the inverse Hessian matrix has cross-platform differences in
# mean relative difference on the order of 1.0e-08, which is not a problem.
# m1r2$NegInvHessian <- round(m1r2$NegInvHessian, 6)
# m1r2_test$NegInvHessian <- round(m1r2_test$NegInvHessian, 6)
# # Similarly for other estimates.
# m1r2$SE$db <- round(m1r2$SE$db, 6)
# m1r2_test$SE$db <- round(m1r2_test$SE$db, 6)
# m1r2$SE$muHat <- round(m1r2$SE$muHat, 6)
# m1r2_test$SE$muHat <- round(m1r2_test$SE$muHat, 6)
# m1r2$SE$alphaHat <- round(m1r2$SE$alphaHat, 6)
# m1r2_test$SE$alphaHat <- round(m1r2_test$SE$alphaHat, 6)
# # m1r2$SE$gammaHat <- round(m1r2$SE$gammaHat, 6)
# # m1r2_test$SE$gammaHat <- round(m1r2_test$SE$gammaHat, 6)
# Conduct separate tests on Hessian and standard errors.
m1r2_NegInvHessian <- m1r2$NegInvHessian
m1r2$NegInvHessian <- NULL
m1r2_test_NegInvHessian <- m1r2_test$NegInvHessian
m1r2_test$NegInvHessian <- NULL
m1r2_SE <- m1r2$SE
m1r2$SE <- NULL
m1r2_test_SE <- m1r2_test$SE
m1r2_test$SE <- NULL
# Accuracy can only be so high on 32-bit platforms.
expect_equal(m1r2_test, m1r2)
expect_equal(m1r2_test_NegInvHessian, m1r2_NegInvHessian, tolerance = 10^(-5))
expect_equal(m1r2_test_SE, m1r2_SE, tolerance = 10^(-5))
# Exact equality (without tolerance) is not expected on all platforms.
skip_on_cran()
expect_equal(m1r2_text, m1r2_text_soln)
# expect_equal(m1r2_test, m1r2)
#
# # Similarly, some numbers in SE different in last decimal place.
# expect_equal(m1r2_text, m1r2_text_soln)
# # expect_equal(m1r2_text[1:85], m1r2_text_soln[1:85])
# # expect_equal(m1r2_text[87:103], m1r2_text_soln[87:103])
# # expect_equal(m1r2_text[105:113], m1r2_text_soln[105:113])
})
test_that("restricted estimation with R_Alpha <- c(0, 1, 0) is correct", {
load(file = 'soln_estn/results_m1r124.RData')
m1r4_text_soln <- readLines('soln_estn/results_m1r4.txt')
opt <- FCVARoptions(
unc_optim_control = list(maxit = 1000, # Reduce tolerance.
reltol = 1e-10),
con_optim_control = list(maxit = 1000, # Reduce tolerance.
pgtol = 1e-10),
hess_delta = 10^(-3), # Stable for testing on 32-bit, 386 architecture.
gridSearch = 0, # Disable grid search in optimization.
dbMin = c(0.01, 0.01), # Set lower bound for d,b.
dbMax = c(2.00, 2.00), # Set upper bound for d,b.
constrained = 0, # Impose restriction dbMax >= d >= b >= dbMin ? 1 <- yes, 0 <- no.
plotRoots = 0 # Don't create plots for tests.
)
x <- votingJNP2014[, c("lib", "ir_can", "un_can")]
opt1 <- opt
opt1$R_Alpha <- matrix(c(0, 1, 0), nrow = 1, ncol = 3)
# opt1$db0 <- c(0.575, 0.575) # Set starting values for optimization algorithm.
capture.output(m1r4_test <- FCVARestn(x, k = 2, r = 1, opt1),
file = 'soln_estn/temp.txt')
m1r4_text <- readLines('soln_estn/temp.txt')
# Note that the inverse Hessian matrix has cross-platform differences in
# mean relative difference on the order of 1.0e-08, which is not a problem.
# m1r4$NegInvHessian <- round(m1r4$NegInvHessian, 6)
# m1r4_test$NegInvHessian <- round(m1r4_test$NegInvHessian, 6)
# # Similarly for other estimates.
# m1r4$SE$db <- round(m1r4$SE$db, 6)
# m1r4_test$SE$db <- round(m1r4_test$SE$db, 6)
# m1r4$SE$muHat <- round(m1r4$SE$muHat, 6)
# m1r4_test$SE$muHat <- round(m1r4_test$SE$muHat, 6)
# m1r4$SE$alphaHat <- round(m1r4$SE$alphaHat, 6)
# m1r4_test$SE$alphaHat <- round(m1r4_test$SE$alphaHat, 6)
# # m1r4$SE$gammaHat <- round(m1r4$SE$gammaHat, 6)
# # m1r4_test$SE$gammaHat <- round(m1r4_test$SE$gammaHat, 6)
# Conduct separate tests on Hessian and standard errors.
m1r4_NegInvHessian <- m1r4$NegInvHessian
m1r4$NegInvHessian <- NULL
m1r4_test_NegInvHessian <- m1r4_test$NegInvHessian
m1r4_test$NegInvHessian <- NULL
m1r4_SE <- m1r4$SE
m1r4$SE <- NULL
m1r4_test_SE <- m1r4_test$SE
m1r4_test$SE <- NULL
# Additional comparisons on M1mac found numerical differences in gammaHat.
m1r4_test_SE_gammaHat <- m1r4_test_SE$gammaHat
m1r4_test_SE$gammaHat <- NULL
m1r4_SE_gammaHat <- m1r4_SE$gammaHat
m1r4_SE$gammaHat <- NULL
# Accuracy can only be so high on 32-bit platforms.
expect_equal(m1r4_test, m1r4)
expect_equal(m1r4_test_NegInvHessian, m1r4_NegInvHessian, tolerance = 10^(-5))
expect_equal(m1r4_test_SE, m1r4_SE, tolerance = 10^(-5))
# expect_equal(m1r4_test_SE_gammaHat, m1r4_SE_gammaHat, tolerance = 10^(-5))
# Exact equality (without tolerance) is not expected on all platforms.
skip_on_cran()
expect_equal(m1r4_text, m1r4_text_soln)
# expect_equal(m1r4_test, m1r4)
#
# # Similarly, some numbers in SE different in last decimal place.
# expect_equal(m1r4_text, m1r4_text_soln)
# # expect_equal(m1r4_text[1:85], m1r4_text_soln[1:85])
# # expect_equal(m1r4_text[87:103], m1r4_text_soln[87:103])
# # expect_equal(m1r4_text[105:113], m1r4_text_soln[105:113])
})
LeeMorinUCF/FCVAR documentation built on May 8, 2022, 3:34 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.
context("Estimation")
test_that("estimation options have correct defaults", {
load(file = 'soln_estn/opt.RData')
expect_equal(FCVARoptions(), opt)
})
test_that("estimation options are updated correctly", {
load(file = 'soln_estn/newOpt.RData')
opt <- FCVARoptions(
gridSearch = 0, # Disable grid search in optimization.
dbMin = c(0.01, 0.01), # Set lower bound for d,b.
dbMax = c(2.00, 2.00), # Set upper bound for d,b.
constrained = 0 # Impose restriction dbMax >= d >= b >= dbMin ? 1 <- yes, 0 <- no.
)
expect_equal(FCVARoptionUpdates(opt, p = 3, r = 1), newOpt)
})
test_that("default bounds on parameter space are correct", {
load(file = 'soln_estn/UB_LB_bounds_def.RData')
opt <- FCVARoptions()
expect_equal(GetBounds(opt), UB_LB_bounds)
})
test_that("modified bounds on parameter space are correct", {
load(file = 'soln_estn/UB_LB_bounds_mod.RData')
opt <- FCVARoptions(
dbMin = c(0.01, 0.01), # Set lower bound for d,b.
dbMax = c(2.00, 2.00) # Set upper bound for d,b.
)
expect_equal(GetBounds(opt), UB_LB_bounds)
})
test_that("unrestricted estimation results and output are correct", {
load(file = 'soln_estn/results_m1.RData')
results_text_soln <- readLines('soln_estn/results_m1.txt')
opt <- FCVARoptions(
unc_optim_control = list(maxit = 1000, # Reduce tolerance.
reltol = 1e-10),
con_optim_control = list(maxit = 1000, # Reduce tolerance.
pgtol = 1e-10),
hess_delta = 10^(-3), # Stable for testing on 32-bit, 386 architecture.
gridSearch = 0, # Disable grid search in optimization.
dbMin = c(0.01, 0.01), # Set lower bound for d,b.
dbMax = c(2.00, 2.00), # Set upper bound for d,b.
constrained = 0, # Impose restriction dbMax >= d >= b >= dbMin ? 1 <- yes, 0 <- no.
plotRoots = 0 # Don't create plots for tests.
)
x <- votingJNP2014[, c("lib", "ir_can", "un_can")]
# results_test <- FCVARestn(x, k = 2, r = 1, opt)
capture.output(results_test <- FCVARestn(x, k = 2, r = 1, opt), file = 'soln_estn/temp.txt')
results_text <- readLines('soln_estn/temp.txt')
# Note that the inverse Hessian matrix has cross-platform differences in
# mean relative difference of 1.968147e-08, which is not a problem.
# results$NegInvHessian <- round(results$NegInvHessian, 6)
# results_test$NegInvHessian <- round(results_test$NegInvHessian, 6)
# # Similarly for other estimates.
# results$SE$db <- round(results$SE$db, 6)
# results_test$SE$db <- round(results_test$SE$db, 6)
# results$SE$muHat <- round(results$SE$muHat, 6)
# results_test$SE$muHat <- round(results_test$SE$muHat, 6)
# results$SE$alphaHat <- round(results$SE$alphaHat, 6)
# results_test$SE$alphaHat <- round(results_test$SE$alphaHat, 6)
# # results$SE$gammaHat <- round(results$SE$gammaHat, 6)
# # results_test$SE$gammaHat <- round(results_test$SE$gammaHat, 6)
# Conduct separate tests on Hessian and standard errors.
results_NegInvHessian <- results$NegInvHessian
results$NegInvHessian <- NULL
results_test_NegInvHessian <- results_test$NegInvHessian
results_test$NegInvHessian <- NULL
results_SE <- results$SE
results$SE <- NULL
results_test_SE <- results_test$SE
results_test$SE <- NULL
# Accuracy can only be so high on 32-bit platforms.
expect_equal(results_test, results)
expect_equal(results_test_NegInvHessian, results_NegInvHessian, tolerance = 10^(-5))
expect_equal(results_test_SE, results_SE, tolerance = 10^(-5))
expect_equal(results_text, results_text_soln)
})
test_that("restricted estimation with d = b = 1 is correct", {
load(file = 'soln_estn/results_m1r124.RData')
m1r1_text_soln <- readLines('soln_estn/results_m1r1.txt')
opt <- FCVARoptions(
unc_optim_control = list(maxit = 1000, # Reduce tolerance.
reltol = 1e-10),
con_optim_control = list(maxit = 1000, # Reduce tolerance.
pgtol = 1e-10),
hess_delta = 10^(-3), # Stable for testing on 32-bit, 386 architecture.
gridSearch = 0, # Disable grid search in optimization.
dbMin = c(0.01, 0.01), # Set lower bound for d,b.
dbMax = c(2.00, 2.00), # Set upper bound for d,b.
constrained = 0, # Impose restriction dbMax >= d >= b >= dbMin ? 1 <- yes, 0 <- no.
plotRoots = 0 # Don't create plots for tests.
)
x <- votingJNP2014[, c("lib", "ir_can", "un_can")]
opt1 <- opt
opt1$R_psi <- matrix(c(1, 0), nrow = 1, ncol = 2)
opt1$r_psi <- 1
capture.output(m1r1_test <- FCVARestn(x, k = 2, r = 1, opt1),
file = 'soln_estn/temp.txt')
m1r1_text <- readLines('soln_estn/temp.txt')
# Note that the inverse Hessian matrix has cross-platform differences in
# mean relative difference on the order of 1.0e-08, which is not a problem.
# m1r1$NegInvHessian <- round(m1r1$NegInvHessian, 6)
# m1r1_test$NegInvHessian <- round(m1r1_test$NegInvHessian, 6)
# # Similarly for other estimates.
# m1r1$SE$db <- round(m1r1$SE$db, 6)
# m1r1_test$SE$db <- round(m1r1_test$SE$db, 6)
# m1r1$SE$muHat <- round(m1r1$SE$muHat, 6)
# m1r1_test$SE$muHat <- round(m1r1_test$SE$muHat, 6)
# m1r1$SE$alphaHat <- round(m1r1$SE$alphaHat, 6)
# m1r1_test$SE$alphaHat <- round(m1r1_test$SE$alphaHat, 6)
# # m1r1$SE$gammaHat <- round(m1r1$SE$gammaHat, 6)
# # m1r1_test$SE$gammaHat <- round(m1r1_test$SE$gammaHat, 6)
# Conduct separate tests on Hessian and standard errors.
m1r1_NegInvHessian <- m1r1$NegInvHessian
m1r1$NegInvHessian <- NULL
m1r1_test_NegInvHessian <- m1r1_test$NegInvHessian
m1r1_test$NegInvHessian <- NULL
m1r1_SE <- m1r1$SE
m1r1$SE <- NULL
m1r1_test_SE <- m1r1_test$SE
m1r1_test$SE <- NULL
# Accuracy can only be so high on 32-bit platforms.
expect_equal(m1r1_test, m1r1)
expect_equal(m1r1_test_NegInvHessian, m1r1_NegInvHessian, tolerance = 10^(-5))
expect_equal(m1r1_test_SE, m1r1_SE, tolerance = 10^(-5))
# Exact equality (without tolerance) is not expected on all platforms.
skip_on_cran()
expect_equal(m1r1_text, m1r1_text_soln)
})
test_that("restricted estimation with R_Beta <- c(1, 0, 0) is correct", {
load(file = 'soln_estn/results_m1r124.RData')
m1r2_text_soln <- readLines('soln_estn/results_m1r2.txt')
opt <- FCVARoptions(
unc_optim_control = list(maxit = 1000, # Reduce tolerance.
reltol = 1e-10),
con_optim_control = list(maxit = 1000, # Reduce tolerance.
pgtol = 1e-10),
hess_delta = 10^(-3), # Stable for testing on 32-bit, 386 architecture.
gridSearch = 0, # Disable grid search in optimization.
dbMin = c(0.01, 0.01), # Set lower bound for d,b.
dbMax = c(2.00, 2.00), # Set upper bound for d,b.
constrained = 0, # Impose restriction dbMax >= d >= b >= dbMin ? 1 <- yes, 0 <- no.
plotRoots = 0 # Don't create plots for tests.
)
x <- votingJNP2014[, c("lib", "ir_can", "un_can")]
opt1 <- opt
opt1$R_Beta <- matrix(c(1, 0, 0), nrow = 1, ncol = 3)
# opt1$db0 <- c(0.67, 0.67) # Set starting values for optimization algorithm.
capture.output(m1r2_test <- FCVARestn(x, k = 2, r = 1, opt1),
file = 'soln_estn/temp.txt')
m1r2_text <- readLines('soln_estn/temp.txt')
# Note that the inverse Hessian matrix has cross-platform differences in
# mean relative difference on the order of 1.0e-08, which is not a problem.
# m1r2$NegInvHessian <- round(m1r2$NegInvHessian, 6)
# m1r2_test$NegInvHessian <- round(m1r2_test$NegInvHessian, 6)
# # Similarly for other estimates.
# m1r2$SE$db <- round(m1r2$SE$db, 6)
# m1r2_test$SE$db <- round(m1r2_test$SE$db, 6)
# m1r2$SE$muHat <- round(m1r2$SE$muHat, 6)
# m1r2_test$SE$muHat <- round(m1r2_test$SE$muHat, 6)
# m1r2$SE$alphaHat <- round(m1r2$SE$alphaHat, 6)
# m1r2_test$SE$alphaHat <- round(m1r2_test$SE$alphaHat, 6)
# # m1r2$SE$gammaHat <- round(m1r2$SE$gammaHat, 6)
# # m1r2_test$SE$gammaHat <- round(m1r2_test$SE$gammaHat, 6)
# Conduct separate tests on Hessian and standard errors.
m1r2_NegInvHessian <- m1r2$NegInvHessian
m1r2$NegInvHessian <- NULL
m1r2_test_NegInvHessian <- m1r2_test$NegInvHessian
m1r2_test$NegInvHessian <- NULL
m1r2_SE <- m1r2$SE
m1r2$SE <- NULL
m1r2_test_SE <- m1r2_test$SE
m1r2_test$SE <- NULL
# Accuracy can only be so high on 32-bit platforms.
expect_equal(m1r2_test, m1r2)
expect_equal(m1r2_test_NegInvHessian, m1r2_NegInvHessian, tolerance = 10^(-5))
expect_equal(m1r2_test_SE, m1r2_SE, tolerance = 10^(-5))
# Exact equality (without tolerance) is not expected on all platforms.
skip_on_cran()
expect_equal(m1r2_text, m1r2_text_soln)
# expect_equal(m1r2_test, m1r2)
#
# # Similarly, some numbers in SE different in last decimal place.
# expect_equal(m1r2_text, m1r2_text_soln)
# # expect_equal(m1r2_text[1:85], m1r2_text_soln[1:85])
# # expect_equal(m1r2_text[87:103], m1r2_text_soln[87:103])
# # expect_equal(m1r2_text[105:113], m1r2_text_soln[105:113])
})
test_that("restricted estimation with R_Alpha <- c(0, 1, 0) is correct", {
load(file = 'soln_estn/results_m1r124.RData')
m1r4_text_soln <- readLines('soln_estn/results_m1r4.txt')
opt <- FCVARoptions(
unc_optim_control = list(maxit = 1000, # Reduce tolerance.
reltol = 1e-10),
con_optim_control = list(maxit = 1000, # Reduce tolerance.
pgtol = 1e-10),
hess_delta = 10^(-3), # Stable for testing on 32-bit, 386 architecture.
gridSearch = 0, # Disable grid search in optimization.
dbMin = c(0.01, 0.01), # Set lower bound for d,b.
dbMax = c(2.00, 2.00), # Set upper bound for d,b.
constrained = 0, # Impose restriction dbMax >= d >= b >= dbMin ? 1 <- yes, 0 <- no.
plotRoots = 0 # Don't create plots for tests.
)
x <- votingJNP2014[, c("lib", "ir_can", "un_can")]
opt1 <- opt
opt1$R_Alpha <- matrix(c(0, 1, 0), nrow = 1, ncol = 3)
# opt1$db0 <- c(0.575, 0.575) # Set starting values for optimization algorithm.
capture.output(m1r4_test <- FCVARestn(x, k = 2, r = 1, opt1),
file = 'soln_estn/temp.txt')
m1r4_text <- readLines('soln_estn/temp.txt')
# Note that the inverse Hessian matrix has cross-platform differences in
# mean relative difference on the order of 1.0e-08, which is not a problem.
# m1r4$NegInvHessian <- round(m1r4$NegInvHessian, 6)
# m1r4_test$NegInvHessian <- round(m1r4_test$NegInvHessian, 6)
# # Similarly for other estimates.
# m1r4$SE$db <- round(m1r4$SE$db, 6)
# m1r4_test$SE$db <- round(m1r4_test$SE$db, 6)
# m1r4$SE$muHat <- round(m1r4$SE$muHat, 6)
# m1r4_test$SE$muHat <- round(m1r4_test$SE$muHat, 6)
# m1r4$SE$alphaHat <- round(m1r4$SE$alphaHat, 6)
# m1r4_test$SE$alphaHat <- round(m1r4_test$SE$alphaHat, 6)
# # m1r4$SE$gammaHat <- round(m1r4$SE$gammaHat, 6)
# # m1r4_test$SE$gammaHat <- round(m1r4_test$SE$gammaHat, 6)
# Conduct separate tests on Hessian and standard errors.
m1r4_NegInvHessian <- m1r4$NegInvHessian
m1r4$NegInvHessian <- NULL
m1r4_test_NegInvHessian <- m1r4_test$NegInvHessian
m1r4_test$NegInvHessian <- NULL
m1r4_SE <- m1r4$SE
m1r4$SE <- NULL
m1r4_test_SE <- m1r4_test$SE
m1r4_test$SE <- NULL
# Additional comparisons on M1mac found numerical differences in gammaHat.
m1r4_test_SE_gammaHat <- m1r4_test_SE$gammaHat
m1r4_test_SE$gammaHat <- NULL
m1r4_SE_gammaHat <- m1r4_SE$gammaHat
m1r4_SE$gammaHat <- NULL
# Accuracy can only be so high on 32-bit platforms.
expect_equal(m1r4_test, m1r4)
expect_equal(m1r4_test_NegInvHessian, m1r4_NegInvHessian, tolerance = 10^(-5))
expect_equal(m1r4_test_SE, m1r4_SE, tolerance = 10^(-5))
# expect_equal(m1r4_test_SE_gammaHat, m1r4_SE_gammaHat, tolerance = 10^(-5))
# Exact equality (without tolerance) is not expected on all platforms.
skip_on_cran()
expect_equal(m1r4_text, m1r4_text_soln)
# expect_equal(m1r4_test, m1r4)
#
# # Similarly, some numbers in SE different in last decimal place.
# expect_equal(m1r4_text, m1r4_text_soln)
# # expect_equal(m1r4_text[1:85], m1r4_text_soln[1:85])
# # expect_equal(m1r4_text[87:103], m1r4_text_soln[87:103])
# # expect_equal(m1r4_text[105:113], m1r4_text_soln[105:113])
})
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.