adds tile plots for classification tasks (#923)

NEWS

@@ -30,6 +30,7 @@ mlr_2.9:
 - rename generatePartialPrediction, plotPartialPrediction, and plotPartialPredictionGGVIS to
   generatePartialDependence, plotPartialDependence, and plotPartialDependenceGGVIS
 - add stability.nugget parameter for "regr.km"
+- plotPartialDependence now can create plots classification tasks with more than one interacted features
 
 - new learners
 -- surv.cv.CoxBoost

R/generatePartialDependence.R

@@ -460,26 +460,8 @@ plotPartialDependence = function(obj, geom = "line", facet = NULL, p = 1) {
   if (length(obj$features) > 2L & geom != "tile" & obj$interaction)
     stop("To plot more than 2 features geom must be 'tile'!")
   assertChoice(geom, c("tile", "line"))
-  if (geom == "tile") {
-    if (!(obj$task.desc$type %in% c("regr", "surv"))) {
-      if (length(obj$task.desc$class.levels) > 2L)
-        stop("Only visualization of binary classification works with tiling!")
-    }
-
-    feat_classes = sapply(obj$data, class)
-    if (any(feat_classes == "factor")) {
-      fact_feats = names(feat_classes[feat_classes == "factor"])
-      if (!is.null(facet))
-        fact_feats = fact_feats[which(fact_feats != facet)]
-      do_not_contour = length(fact_feats) > 0L
-    } else
-      do_not_contour = FALSE
-
-    if (do_not_contour)
-      warning("Factor features cannot be used to create contour plots! only tiles will be displayed.")
-    if (!obj$interaction)
+  if (geom == "tile" & !obj$interaction)
       stop("generatePartialDependenceData was called with interaction = FALSE!")
-  }
 
   if (!is.null(facet)) {
     assertChoice(facet, obj$features)
@@ -499,6 +481,8 @@ plotPartialDependence = function(obj, geom = "line", facet = NULL, p = 1) {
       facet = "Feature"
       scales = "free_x"
     }
+    if (obj$task.desc$type == "classif" & geom == "tile" & length(features) == 2L)
+      scales = "free_x"
   }
 
   if (p != 1) {
@@ -554,14 +538,18 @@ plotPartialDependence = function(obj, geom = "line", facet = NULL, p = 1) {
     if (obj$derivative)
       plt = plt + ylab(stri_paste(target, "(derivative)", sep = " "))
   } else { ## tiling
-    plt = ggplot(obj$data, aes_string(x = features[1], y = features[2], z = target))
-    plt = plt + geom_tile(aes_string(fill = target))
-    if (!do_not_contour)
-      plt = plt + stat_contour()
+    if (obj$task.desc$type == "classif") {
+      plt = ggplot(obj$data, aes_string(x = features[1], y = features[2], fill = "Probability"))
+      plt = plt + geom_raster()
+      facet = "Class"
+    } else {
+      plt = ggplot(obj$data, aes_string(x = features[1], y = features[2], z = target))
+      plt = plt + geom_raster(aes_string(fill = target))
+    }
   }
 
   if (!is.null(facet))
-    plt = plt + facet_wrap(as.formula(stri_paste("~ ", facet)), scales = scales)
+    plt = plt + facet_wrap(as.formula(stri_paste("~", facet)), scales = scales)
 
   plt
 }

tests/testthat/test_base_generatePartialDependence.R

@@ -72,7 +72,7 @@ test_that("generatePartialDependenceData", {
                                       interaction = TRUE, gridsize = gridsize)
   nfacet = length(unique(dcp$data$Petal.Length))
   ntarget = length(dcp$target)
-  plotPartialDependence(dcp, facet = "Petal.Length")
+  plotPartialDependence(dcp, geom = "tile")
   ggsave(path)
   doc = XML::xmlParse(path)
   #expect_that(length(XML::getNodeSet(doc, grey.xpath, ns.svg)), equals(nfacet))
@@ -84,11 +84,11 @@ test_that("generatePartialDependenceData", {
   ## check that probability outputting classifiers work with ICE
   dcp = generatePartialDependenceData(fcp, input = multiclass.task, features = c("Petal.Width", "Petal.Length"),
                                       interaction = TRUE, individual = TRUE, gridsize = gridsize)
-  plotPartialDependence(dcp, facet = "Petal.Length")
+  plotPartialDependence(dcp, geom = "tile")
   ## plotPartialDependenceGGVIS(dcp, interact = "Petal.Length")
 
   ## check that survival tasks work with multiple features
-  fs = train("surv.rpart", surv.task)
+  fs = train("surv.coxph", surv.task)
   ds = generatePartialDependenceData(fs, input = surv.task, features = c("x1", "x2"),
                                      gridsize = gridsize)
   nfeat = length(ds$features)
@@ -167,8 +167,6 @@ test_that("generatePartialDependenceData", {
                                        bounds = c(-2, 2), gridsize = gridsize)
 
   ## check that tile + contour plots work for two and three features with regression and survival
-  expect_warning(plotPartialDependence(db, "tile")) ## factor feature
-  expect_error(plotPartialDependence(dcp, geom = "tile")) ## no multiclass support
   expect_error(plotPartialDependence(ds, geom = "tile")) ## interaction == FALSE
   tfr = generatePartialDependenceData(fr, regr.df, features = c("lstat", "crim", "chas"),
                                       interaction = TRUE, gridsize = gridsize)