tests/testthat/test-nlpMarginal-groups.R

In mombf: Model Selection with Bayesian Methods and Information Criteria

context("Test nlpMarginal with groups")
library("mombf")

source(test_path("data-for-tests.R"))
tolerance <- 1e-6

patrick::with_parameters_test_that(
  "check_sel_groups works", {
    groups <- c(1,2,2,2,3,3,4,5)
    expect_error(check_sel_groups(sel, groups), error_msg)
  },
  patrick::cases(
    sel_outofbounds=list(sel=c(1,7,20), error_msg="sel larger than"),
    sel_incompatible_groups=list(sel=c(1,2,7), error_msg="incompatible with .+ groups"),
    sel_good_int=list(sel=c(2,3,4,7,8), error_msg=NA),
    sel_good_bool=list(sel=c(TRUE,FALSE,FALSE,FALSE,TRUE,TRUE,TRUE, FALSE), error_msg=NA)
  )
)

test_that(
  "nlpMarginal ignores sel argument when its a formula", {
    pCoef <- momprior(tau=0.348)
    ans1 <- nlpMarginal(seq(4), y3, X3, priorCoef=pCoef)
    formula <- y3~X3[,2]+X3[,3]+X3[,4]
    expect_warning(ans2 <- nlpMarginal(c(1,2), y=formula, priorCoef=pCoef), "ignoring sel")
    expect_equal(ans1, ans2)
    expect_warning(ans3 <- nlpMarginal(y=formula, priorCoef=pCoef), NA)
    expect_equal(ans1, ans3)
  }
)

patrick::with_parameters_test_that(
  "nlpMarginal with same kind of priorCoef and Group is correctly impemented for normal family:", {
    pVar <- igprior(alpha=0.01, lambda=0.01)
    ans_max <- nlpMarginal(theta9_truth_idx, y9, X9, family="normal", priorCoef=pCoef, priorVar=pVar)
    ans_max_group <- nlpMarginal(
      theta9_truth_idx, y9, X9, groups=groups9, family="normal",
      priorCoef=pCoef, priorGroup=pCoef, priorVar=pVar
    )

    ans_all <- nlpMarginal(seq_along(theta9_truth), y9, X9, family="normal",priorCoef=pCoef, priorVar=pVar)
    ans_all_group <- nlpMarginal(
      seq_along(theta9_truth), y9, X9, groups=groups9, family="normal",
      priorCoef=pCoef, priorGroup=pCoef, priorVar=pVar
    )
    expect_true(ans_max_group > ans_all_group)
    expect_equal(ans_max, ans_max_group)
    expect_equal(ans_all, ans_all_group)
  },
  test_name=c("mom", "imom", "emom", "zellner", "normalid"),
  pCoef=c(momprior(tau=0.35), imomprior(tau=0.35), emomprior(tau=0.35), zellnerprior(tau=0.35), normalidprior(tau=0.35))
)

patrick::with_parameters_test_that(
  "nlpMarginal with groups is correctly impemented for normal family:", {
    pVar <- igprior(alpha=0.01, lambda=0.01)
    ans_max <- nlpMarginal(
      theta9_truth_idx, y9, X9, groups=groups9, family="normal",
      priorCoef=pCoef, priorGroup=pGroup, priorVar=pVar
    )

    ans_all <- nlpMarginal(
      seq_along(theta9_truth), y9, X9, groups=groups9, family="normal",
      priorCoef=pCoef, priorGroup=pGroup, priorVar=pVar
    )
    expect_equal(ans_max, expected_max, tolerance=tolerance)
    expect_equal(ans_all, expected_all, tolerance=tolerance)
  },
  patrick::cases(
    normalid=list(pCoef=normalidprior(tau=0.3), pGroup=normalidprior(tau=0.4), expected_max=-231.111757, expected_all=-236.971966),
    normid_gzell=list(pCoef=normalidprior(tau=1), pGroup=groupzellnerprior(tau=0.4), expected_max=-316.658333, expected_all=-320.087106),
    zell_gzell=list(pCoef=zellnerprior(tau=0.3), pGroup=groupzellnerprior(tau=0.4), expected_max=-342.339794, expected_all=-342.19151),
    mom_gzell=list(pCoef=momprior(tau=0.3), pGroup=groupzellnerprior(tau=10), expected_max=-262.450681, expected_all=-274.104395),
    mom_gmom=list(pCoef=momprior(tau=0.3), pGroup=groupmomprior(tau=0.4), expected_max=-238.026357, expected_all=-252.308801)
  )
)