added first tests for PipeOpClassWeightsEx

Author: awinterstetter

R/PipeOpClassWeights.R

@@ -97,7 +97,7 @@ PipeOpClassWeights = R6Class("PipeOpClassWeights",
   ),
   private = list(
     .train_task = function(task) {
-      browser()
+
       pv = self$param_set$get_values(tags = "train")
 
       if ("twoclass" %nin% task$properties) {

R/PipeOpClassWeightsEx.R

@@ -101,17 +101,19 @@ PipeOpClassWeightsEx = R6Class("PipeOpClassWeightsEx",
     .train_task = function(task) {
       pv = self$param_set$get_values(tags = "train")
 
+      if (is.null(pv$weight_type) ||
+          is.null(pv$weight_method) ||
+          (pv$weight_method == "explicit" && is.null(pv$mapping))) {
+        return(task)
+      }
+
       weightcolname = ".WEIGHTS"
       if (weightcolname %in% unlist(task$col_roles)) {
         stopf("Weight column '%s' is already in the Task", weightcolname)
       }
 
       truth = task$truth()
 
-      if (is.null(pv$weight_type)) {
-        return(task)
-      }
-
       class_frequency = table(truth) / length(truth)
       class_names = names(class_frequency)
 
@@ -122,7 +124,8 @@ PipeOpClassWeightsEx = R6Class("PipeOpClassWeightsEx",
         "explicit" = pv$mapping
       )
 
-      wcol = setnames(data.table(weights_by_class[truth])[, "N"], weightcolname)
+      weights_table = data.table(weights_by_class[truth])
+      wcol = setnames(as.data.table(weights_table[[ncol(weights_table)]]), weightcolname)
       task$cbind(wcol)
       task$col_roles$feature = setdiff(task$col_roles$feature, weightcolname)
 

tests/testthat/test_pipeop_classweightsex.R

@@ -0,0 +1,50 @@
+
+test_that("PipeOpClassWeightsEx - basic properties", {
+  op = PipeOpClassWeightsEx$new()
+  task = mlr_tasks$get("german_credit")
+  expect_pipeop(op)
+  train_pipeop(op, inputs = list(task))
+  predict_pipeop(op, inputs = list(task))
+
+  expect_datapreproc_pipeop_class(PipeOpClassWeights, task = task,
+                                  predict_like_train = FALSE)
+})
+
+test_that("PipeOpClassWeightsEx - error for Tasks without weights property, #937", {
+  skip_if_not_installed("mlr3learners")
+  skip_if_not_installed("MASS")
+
+  set.seed(1234)
+  task = as_task_classif(data.table(
+    y = factor(rep(c("A", "B", "A", "C"), 4)),
+    x = runif(16)
+  ), target = "y")
+
+  # no error: Learner has weights property
+  gr = po("classweightsex", param_vals = list(mapping = c("A" = 0.6, "B" = 0.3, "C" = 0.1))) %>>% lrn("classif.featureless")
+  expect_no_error(gr$train(task))
+
+  # error: Learner does not have weights property
+  gr = po("classweightsex",  param_vals = list(mapping = c("A" = 0.6, "B" = 0.3, "C" = 0.1))) %>>% lrn("classif.lda")
+  expect_error(gr$train(task), ".*Learner does not support weights.*")
+
+  # no error: use_weights is set to "ignore"
+  gr = po("classweightsex", param_vals = list(mapping = c("A" = 0.6, "B" = 0.4, "C" = 0.1))) %>>% lrn("classif.lda", use_weights = "ignore")
+  expect_no_error(gr$train(task))
+
+})
+
+test_that("PipeOpClassWeightsEx", {
+
+  task = mlr_tasks$get("iris")
+
+  # Method inverse_class_frequency
+  poicf = po("classweightsex", param_vals = list(weight_method = "inverse_class_frequency"))
+  nt = poicf$train(list(task))[[1L]]
+  expect_equal(nt$data(), task$data())
+
+
+  # manual_weights = as.data.table(1 / table(task$data()$Species))
+  # weights = if ("weights_learner" %in% names(nt)) "weights_learner" else "weights"
+  # expect_equal(nt[[weights]]$weight, ifelse(nt$truth(nt[[weights]]$row_ids) == "neg", 1, 3))
+})