Download Latest Version mlr3 1.8.0 source code.zip (1.4 MB) Google Add to Preferred Sources
Home / v1.7.0
Name Modified Size InfoDownloads / Week
Parent folder
mlr3 1.7.0 source code.tar.gz 2026-06-09 928.5 kB
0
mlr3 1.7.0 source code.zip 2026-06-09 1.2 MB
0
README.md 2026-06-09 2.2 kB
0
Totals: 3 Items   2.2 MB 0
  • BREAKING CHANGE: Removed the deprecated Task$divide() method, use the $internal_valid_task field instead.
  • feat: Encapsulated Learners now support a deadline (date-time) in addition to the existing timeout (seconds).
  • perf: Vectorized the row-wise NA check in the prediction fallback path.
  • refactor: Removed unused task_prototype from the learner state.
  • fix: as_measures() now forwards task_type when converting a list.
  • fix: as_task_classif() and as_task_regr() now respect the drop_original_target and drop_levels arguments when converting a task instead of ignoring them.
  • fix: as_tasks_unsupervised() now correctly converts lists to TaskUnsuperviseds.
  • fix: c() on regression Predictions no longer drops the probs attribute of the combined quantiles matrix.
  • fix: importance() and selected_features() now raise a learner error instead of an input error when the model is not trained yet.
  • fix: Measure$obs_loss() now forwards the learner argument to the private .obs_loss() method instead of dropping it.
  • fix: Measure$score() no longer errors during the task_type check when prediction is NULL.
  • fix: Reset range of MeasureClassifCost.
  • fix: MeasureClassifSimple and MeasureRegrSimple now forward their hyperparameter values to the underlying scoring function instead of ignoring them.
  • fix: mlr_measures_regr.aic now accepts non-integer values for k by changing the parameter from an integer to a double.
  • fix: MeasureRegrPinball now declares the "weights" property and forwards sample weights to the underlying scoring function.
  • fix: MeasureSimilarity$new() no longer accepts the silently ignored average argument.
  • fix: ResamplingCustom and ResamplingCustomCV now invalidate the cached $hash on re-instantiation.
  • fix: Task$hash is now read-only.
  • fix: TaskClassif$droplevels() now forwards the cols argument to the parent method instead of dropping levels on all columns.
  • fix: The predict path now raises a learner-predict error instead of a learner-train error.
  • fix: Validation of the weights columns now uses the backend row names instead of a row sequence, fixing checks on tasks with non-contiguous row ids.
Source: README.md, updated 2026-06-09