tests/testthat/test_base_NoFeaturesModel.R

In mlr: Machine Learning in R

test_that("NoFeaturesModel", {
  task = subsetTask(multiclass.task, features = character(0))
  lrn = makeLearner("classif.lda", predict.type = "prob")
  m = train(lrn, task)
  expect_s3_class(m$learner.model, "NoFeaturesModel")
  expect_s3_class(m$learner, "classif.lda")
  expect_equal(m$learner$predict.type, "prob")

  p = predict(m, newdata = multiclass.df)
  expect_true(setequal(colnames(as.data.frame(p)),
    c("prob.setosa", "prob.virginica", "prob.versicolor", "truth", "response")))

  res = makeResampleDesc("CV", iter = 2)
  rf = resample(lrn, task, res)
  expect_true(setequal(colnames(as.data.frame(p)),
    c("prob.setosa", "prob.virginica", "prob.versicolor", "truth", "response")))

  task = subsetTask(regr.task, features = character(0))
  lrn = makeLearner("regr.lm")
  m = train(lrn, task)
  p = predict(m, newdata = regr.df)
  expect_true(all(p$data$response == mean(p$data$response)))

  rf = resample(lrn, task, res)$pred
  expect_equal(length(unique(rf$data$response)), 2)
})

test_that("NoFeaturesModel works with FilterWrapper", {
  lrn = makeLearner("classif.rpart")
  lrn = makeFilterWrapper(lrn, fw.method = "anova.test", fw.perc = 0.1)
  m = train(lrn, multiclass.task)
  p = predict(m, multiclass.task)
  expect_true(!is.na(performance(p, measures = getDefaultMeasure(multiclass.task))))
})