Merge branch 'main' of github.com:mlr-org/mlr3viz

Author: be-marc

DESCRIPTION

@@ -49,6 +49,7 @@ Suggests:
     mlr3cluster,
     mlr3filters,
     mlr3fselect (>= 1.0.0),
+	mlr3inference,
     mlr3learners,
     mlr3tuning (>= 1.0.0),
     paradox,
@@ -60,13 +61,20 @@ Suggests:
     stats,
     testthat (>= 3.0.0),
     vdiffr (>= 1.0.2),
-    xgboost
+    xgboost,
+    survminer,
+    mlr3proba (>= 0.6.3)
+Remotes:
+    mlr-org/mlr3proba,
+    mlr-org/mlr3inference
+Additional_repositories:
+  https://mlr-org.r-universe.dev
 Config/testthat/edition: 3
 Config/testthat/parallel: true
 Encoding: UTF-8
 NeedsCompilation: no
 Roxygen: list(markdown = TRUE)
-RoxygenNote: 7.3.1
+RoxygenNote: 7.3.2
 Collate:
     'BenchmarkResult.R'
     'Filter.R'
@@ -79,6 +87,7 @@ Collate:
     'LearnerRegrCVGlmnet.R'
     'LearnerRegrGlmnet.R'
     'LearnerRegrRpart.R'
+    'LearnerSurvCoxPH.R'
     'OptimInstanceBatchSingleCrit.R'
     'Prediction.R'
     'PredictionClassif.R'

NAMESPACE

@@ -18,6 +18,7 @@ S3method(autoplot,LearnerRegr)
 S3method(autoplot,LearnerRegrCVGlmnet)
 S3method(autoplot,LearnerRegrGlmnet)
 S3method(autoplot,LearnerRegrRpart)
+S3method(autoplot,LearnerSurvCoxPH)
 S3method(autoplot,OptimInstanceBatchSingleCrit)
 S3method(autoplot,PredictionClassif)
 S3method(autoplot,PredictionClust)
@@ -41,6 +42,7 @@ S3method(plot,LearnerClassifRpart)
 S3method(plot,LearnerRegrCVGlmnet)
 S3method(plot,LearnerRegrGlmnet)
 S3method(plot,LearnerRegrRpart)
+S3method(plot,LearnerSurvCoxPH)
 S3method(plot,PredictionClassif)
 S3method(plot,PredictionRegr)
 S3method(plot,ResampleResult)

NEWS.md

@@ -1,9 +1,11 @@
 # mlr3viz (development version)
 
+- Add plot for `LearnerSurvCoxPH`.
+- Add plot for confidence intervals (`mlr3inference`)
+
 # mlr3viz 0.9.0
 
 - Work with new bbotk 0.9.0 and mlr3tuning 0.21.0
-
 - Add plots for `EnsembleFSResult` object.
 
 # mlr3viz 0.8.0

R/BenchmarkResult.R

@@ -12,6 +12,7 @@
 #'   Requires package \CRANpkg{precrec}.
 #' * `"prc"`: Precision recall curve.
 #'    See `"roc"`.
+#' * `"ci"`: Plot confidence intervals. Pass a `msr("ci", ...)` from the `mlr3inference` package as argument `measure`.
 #'
 #' @param object ([mlr3::BenchmarkResult]).
 #' @template param_type
@@ -45,6 +46,43 @@ autoplot.BenchmarkResult = function(object, type = "boxplot", measure = NULL, th
 
   task = object$tasks$task[[1L]]
   measure = mlr3::assert_measure(mlr3::as_measure(measure, task_type = task$task_type), task = task)
+
+  if (identical(type, "ci")) {
+    mlr3misc::require_namespaces("mlr3inference")
+
+    assert_class(measure, "MeasureAbstractCi")
+    mid = measure$id
+
+    tbl = object$aggregate(measure)
+
+    tmp = map(object$resamplings$resampling, function(x) {
+      list(class(x), x$param_set$values)
+    })
+
+    if (length(unique(tmp)) != 1) {
+      stopf("Plot of type 'ci' requires exactly one resampling method")
+    }
+
+    # static checker
+    .data = NULL
+    task_id = NULL
+    p = ggplot(tbl, aes(x = .data[["learner_id"]], y = .data[[mid]])) +
+      geom_point() +
+      geom_errorbar(aes(ymin = .data[[paste0(mid, ".lower")]], ymax = .data[[paste0(mid, ".upper")]]), width = 0.2) +
+      facet_wrap(vars(task_id), scales = "free_y") +
+      labs(
+      title = sprintf("Confidence Intervals for alpha = %s", measure$param_set$values$alpha),
+      x = "Learner",
+      y = paste0(measure$measure$id)
+      ) +
+      theme +
+      theme(
+        axis.text.x = element_text(angle = 45, hjust = 1),
+        axis.title.x = element_blank()
+      )
+    return(p)
+  }
+
   measure_id = measure$id
   tab = fortify(object, measure = measure)
   tab$nr = sprintf("%09d", tab$nr)

R/LearnerSurvCoxPH.R

@@ -0,0 +1,49 @@
+#' @title Plots for Cox Proportional Hazards Learner
+#'
+#' @description
+#' Visualizations for [mlr3proba::LearnerSurvCoxPH].
+#'
+#' The argument `type` controls what kind of plot is drawn.
+#' The only possible choice right now is `"ggforest"` (default) which is a
+#' Forest Plot, using [ggforest][survminer::ggforest()].
+#' This plot displays the estimated hazard ratios (HRs) and their confidence
+#' intervals (CIs) for different variables included in the (trained) model.
+#'
+#' @param object ([mlr3proba::LearnerSurvCoxPH]).
+#'
+#' @template param_type
+#' @param ... Additional parameters passed down to `ggforest`.
+#'
+#' @return [ggplot2::ggplot()].
+#'
+#' @export
+#' @examples
+#' \donttest{
+#' if (requireNamespace("mlr3proba")) {
+#'   library(mlr3proba)
+#'   library(mlr3viz)
+#'
+#'   task = tsk("lung")
+#'   learner = lrn("surv.coxph")
+#'   learner$train(task)
+#'   autoplot(learner)
+#' }
+#' }
+autoplot.LearnerSurvCoxPH = function(object, type = "ggforest", ...) {
+  assert_class(object, classes = "LearnerSurvCoxPH", null.ok = FALSE)
+  assert_has_model(object)
+
+  switch(type,
+    "ggforest" = {
+      require_namespaces("survminer")
+      suppressWarnings(survminer::ggforest(object$model, ...))
+     },
+
+     stopf("Unknown plot type '%s'", type)
+  )
+}
+
+#' @export
+plot.LearnerSurvCoxPH = function(x, ...) {
+  print(autoplot(x, ...))
+}

man/autoplot.BenchmarkResult.Rd

@@ -48,3 +48,19 @@ test_that("holdout roc plot (#54)", {
 
   expect_doppelganger("bmr_holdout_roc", p)
 })
+
+skip_if_not_installed("mlr3inference")
+skip_if_not_installed("rpart")
+
+test_that("CI plot", {
+  bmr = benchmark(benchmark_grid(tsks(c("mtcars", "boston_housing")),
+    lrns(c("regr.featureless", "regr.rpart")), rsmp("holdout")))
+
+  p = autoplot(bmr, "ci", msr("ci", "regr.mse"))
+  expect_true(is.ggplot(p))
+  expect_doppelganger("bmr_holdout_ci", p)
+
+  bmr = benchmark(benchmark_grid(tsk("iris"), lrn("classif.rpart"),
+    rsmps(c("holdout", "cv"))))
+  expect_error(autoplot(bmr, "ci", msr("ci", "classif.acc")), "one resampling method")
+})