From 0b1cb8a4a1f88c5ebb6da2b374027688815b7bdd Mon Sep 17 00:00:00 2001
From: Michel Lang <michellang@gmail.com>
Date: Wed, 10 Aug 2022 14:37:52 +0200
Subject: [PATCH] refactor: custom cost measure does not require the task
 anymore

---
 R/MeasureClassifCosts.R           | 29 ++++++++++-------------------
 man/mlr_measures_classif.costs.Rd |  2 --
 2 files changed, 10 insertions(+), 21 deletions(-)

diff --git a/R/MeasureClassifCosts.R b/R/MeasureClassifCosts.R
index 75f832759..e79eb9c7b 100644
--- a/R/MeasureClassifCosts.R
+++ b/R/MeasureClassifCosts.R
@@ -11,8 +11,6 @@
 #' For calculation of the score, the confusion matrix is multiplied element-wise with the cost matrix.
 #' The costs are then summed up (and potentially divided by the number of observations if `normalize` is set to `TRUE` (default)).
 #'
-#' This measure requires the [Task] during scoring to ensure that the rows and columns of the cost matrix are in the same order as in the confusion matrix.
-#'
 #' @templateVar id classif.costs
 #' @template measure
 #'
@@ -52,7 +50,6 @@ MeasureClassifCosts = R6Class("MeasureClassifCosts",
       super$initialize(
         id = "classif.costs",
         param_set = param_set,
-        properties = "requires_task",
         range = c(-Inf, Inf),
         minimize = TRUE,
         label = "Cost-sensitive Classification",
@@ -68,7 +65,10 @@ MeasureClassifCosts = R6Class("MeasureClassifCosts",
       if (missing(rhs)) {
         return(private$.costs)
       }
-      private$.costs = assert_cost_matrix(rhs)
+
+      assert_matrix(rhs, mode = "numeric", any.missing = FALSE, col.names = "unique", row.names = "unique")
+      assert_set_equal(rownames(rhs), colnames(rhs))
+      private$.costs = rhs
 
       if (min(rhs) >= 0) {
         self$range[1L] = 0
@@ -82,8 +82,11 @@ MeasureClassifCosts = R6Class("MeasureClassifCosts",
   private = list(
     .costs = NULL,
 
-    .score = function(prediction, task, ...) {
-      costs = assert_cost_matrix(private$.costs, task)
+    .score = function(prediction, ...) {
+      costs = self$costs
+      lvls = levels(prediction$truth)
+      assert_set_equal(lvls, colnames(costs))
+
       confusion = table(response = prediction$response, truth = prediction$truth, useNA = "ifany")
 
       # reorder rows / cols if necessary
@@ -97,6 +100,7 @@ MeasureClassifCosts = R6Class("MeasureClassifCosts",
       if (self$param_set$values$normalize) {
         perf = perf / sum(confusion)
       }
+
       perf
     },
 
@@ -106,16 +110,3 @@ MeasureClassifCosts = R6Class("MeasureClassifCosts",
 
 #' @include mlr_measures.R
 mlr_measures$add("classif.costs", function() MeasureClassifCosts$new())
-
-assert_cost_matrix = function(costs, task = NULL) {
-  if (is.null(task)) {
-    assert_matrix(costs, mode = "numeric", any.missing = FALSE, col.names = "unique", row.names = "unique")
-  } else {
-    lvls = task$class_names
-    assert_matrix(costs, mode = "numeric", any.missing = FALSE, nrows = length(lvls), ncols = length(lvls), row.names = "unique", col.names = "unique")
-    assert_names(colnames(costs), permutation.of = lvls)
-    assert_names(rownames(costs), permutation.of = lvls)
-  }
-
-  costs
-}
diff --git a/man/mlr_measures_classif.costs.Rd b/man/mlr_measures_classif.costs.Rd
index 7a1652b43..c9383906b 100644
--- a/man/mlr_measures_classif.costs.Rd
+++ b/man/mlr_measures_classif.costs.Rd
@@ -11,8 +11,6 @@ The cost matrix is stored as slot \verb{$costs}.
 
 For calculation of the score, the confusion matrix is multiplied element-wise with the cost matrix.
 The costs are then summed up (and potentially divided by the number of observations if \code{normalize} is set to \code{TRUE} (default)).
-
-This measure requires the \link{Task} during scoring to ensure that the rows and columns of the cost matrix are in the same order as in the confusion matrix.
 }
 \section{Dictionary}{