* Return a long DF for getFeatureImportance()

DESCRIPTION

@@ -264,6 +264,7 @@ Suggests:
     testthat,
     tgp,
     TH.data,
+    tidyr,
     tsfeatures,
     vdiffr,
     wavelets,

NEWS.md

@@ -37,6 +37,8 @@ PR: #2638 (@pfistl)
 ## learners - general
 
 - xgboost: added options 'auto', 'approx' and 'gpu_hist' to param `tree_method` (@albersonmiranda, #2701)
+- `getFeatureImportance()` now returns a long data.frame with columns `variable` and `importance`.
+  Beforehand, a wide data.frame was returned with each variable representing a column (@pat-s, #1755).
 
 ## filters - general
 

R/getFeatureImportance.R

@@ -76,7 +76,8 @@ getFeatureImportance = function(object, ...) {
   # convert named vector to data.frame with columns and set NA to 0
   imp[is.na(imp)] = 0L
   imp = as.data.frame(t(imp))
-  rownames(imp) = NULL
+  imp = tidyr::pivot_longer(imp, tidyr::everything(),
+    names_to = "variable", values_to = "importance")
 
   makeS3Obj("FeatureImportance",
     res = imp,

tests/testthat/helper_learners_all.R

@@ -215,9 +215,11 @@ testThatLearnerCanCalculateImportance = function(lrn, task, hyperpars) {
 
   mod = train(lrn, task)
   feat.imp = getFeatureImportance(mod)$res
-  expect_data_frame(feat.imp, types = rep("numeric", getTaskNFeats(task)),
-    any.missing = FALSE, nrows = 1, ncols = getTaskNFeats(task))
-  expect_equal(colnames(feat.imp), mod$features)
+  expect_data_frame(feat.imp,
+    types = c("character", "numeric"),
+    any.missing = FALSE, nrows = getTaskNFeats(task),
+    ncols = 2)
+  expect_equal(colnames(feat.imp), c("variable", "importance"))
 
 }
 

tests/testthat/test_base_getFeatureImportance.R

@@ -6,36 +6,36 @@ test_that("getFeatureImportance", {
   lrn = makeLearner("classif.randomForest")
   mod = train(lrn, binaryclass.task)
   feat.imp = getFeatureImportance(mod, type = 2)$res
-  expect_data_frame(feat.imp, types = rep("numeric", getTaskNFeats(binaryclass.task)),
-    any.missing = FALSE, nrows = 1, ncols = getTaskNFeats(binaryclass.task))
-  expect_equal(colnames(feat.imp), mod$features)
+  expect_data_frame(feat.imp, types = c("character", "numeric"),
+    any.missing = FALSE, nrows = 60, ncols = 2)
+  expect_equal(colnames(feat.imp), c("variable", "importance"))
 
   # type 1 shouldn't
   expect_error(getFeatureImportance(mod, type = 1), regexp = ".*importance.*TRUE")
 
   lrn = setHyperPars(lrn, importance = TRUE)
   mod = train(lrn, binaryclass.task)
   feat.imp = getFeatureImportance(mod, type = 1)$res
-  expect_data_frame(feat.imp, types = rep("numeric", getTaskNFeats(binaryclass.task)),
-    any.missing = FALSE, nrows = 1, ncols = getTaskNFeats(binaryclass.task))
-  expect_equal(colnames(feat.imp), mod$features)
+  expect_data_frame(feat.imp, types = c("character", "numeric"),
+    any.missing = FALSE, nrows = 60, ncols = 2)
+  expect_equal(colnames(feat.imp), c("variable", "importance"))
 
   # regression learner
   lrn = makeLearner("regr.gbm")
   mod = train(lrn, regr.task)
   feat.imp = getFeatureImportance(mod)$res
-  expect_data_frame(feat.imp, types = rep("numeric", getTaskNFeats(regr.task)),
-    any.missing = FALSE, nrows = 1, ncols = getTaskNFeats(regr.task))
-  expect_equal(colnames(feat.imp), mod$features)
+  expect_data_frame(feat.imp, types = c("character", "numeric"),
+    any.missing = FALSE, nrows = 13, ncols = 2)
+  expect_equal(colnames(feat.imp), c("variable", "importance"))
 
   # wrapped learner
-  lrn = makeFilterWrapper(makeLearner("regr.gbm"), fw.method = "FSelectorRcpp_information.gain", fw.abs = 2,
-    equal = TRUE)
+  lrn = makeFilterWrapper(makeLearner("regr.gbm"),
+    fw.method = "FSelectorRcpp_information.gain", fw.abs = 2, equal = TRUE)
   mod = train(lrn, regr.task)
   feat.imp = getFeatureImportance(mod)$res
-  expect_data_frame(feat.imp, types = rep("numeric", getTaskNFeats(regr.task)),
-    any.missing = FALSE, nrows = 1, ncols = getTaskNFeats(regr.task))
-  expect_equal(colnames(feat.imp), mod$features)
+  expect_data_frame(feat.imp, types = c("character", "numeric"),
+    any.missing = FALSE, nrows = 13, ncols = 2)
+  expect_equal(colnames(feat.imp), c("variable", "importance"))
 
   # For learners without the possibility to calculate feature importance a
   # meaningful error should be returned

tests/testthat/test_featsel_praznik.R

@@ -67,7 +67,8 @@ test_that("FilterWrapper with praznik mutual information, resample", {
   mod = train(lrn, binaryclass.task)
   feat.imp = getFeatureImportance(mod)$res
   expect_data_frame(feat.imp,
-    types = rep("numeric", getTaskNFeats(binaryclass.task)),
-    any.missing = FALSE, nrows = 1, ncols = getTaskNFeats(binaryclass.task))
-  expect_equal(colnames(feat.imp), mod$features)
+    types = c("character", "numeric"),
+    any.missing = FALSE, nrows = getTaskNFeats(binaryclass.task),
+    ncols = 2)
+  expect_equal(colnames(feat.imp), c("variable", "importance"))
 })