tests/testthat/test_featureImportanceLearner.R

In giuseppec/featureImportance: Model-agnostic feature importance with the mlr package

context("featureImportanceLearner works")
test_that("featureImportanceLearner works", {
  n.feat.perm = 2
  feat = list(features[1:2], features[3])

  resampling = list(
    makeResampleInstance(makeResampleDesc("CV", iters = 2), task),
    makeResampleInstance(makeResampleDesc("Bootstrap", iters = 2), task),
    makeResampleInstance(makeResampleDesc("RepCV", folds = 2, reps = 2), task)
  )

  # rin = resampling[[1]]

  for (rin in resampling) {
    imp = featureImportanceLearner(learner, task, rin, features = feat, n.feat.perm = n.feat.perm, measures = measures, local = FALSE)
    nrow = #rin$desc$iters*
      length(feat)*n.feat.perm

    expect_data_table(imp$importance, nrows = nrow)
    expect_subset(c("features", "n.feat.perm", mid), colnames(imp$importance))

    imp.local = featureImportanceLearner(learner, task, rin, features = feat, n.feat.perm = n.feat.perm, measures = measures, local = TRUE)
    res = imp.local$resample
    imp.local = imp.local$importance

    nrow = length(feat)*length(unique(unlist(rin$test.inds)))*n.feat.perm

    expect_data_table(imp.local, nrows = nrow)
    expect_equal(imp.local$acc, -imp.local$mmce)
    expect_equal(stri_split_fixed(unique(imp.local$features), ","), feat)
    expect_subset(c("features", "n.feat.perm", "row.id", mid), colnames(imp.local))
    expect_set_equal(res$pred$data$id, imp.local$row.id)
  }
})