mlr 2.9: 2016-08-03

functions - general

  • various cleanups that removed unused code
  • subsetTask, getTaskData: arg “features” now also accepts logical and integer
  • removeConstantFeatures now also operates on data.frames and makeRemoveConstantFeaturesWrapper can be used to augment a learner with this preprocessing step.
  • normalizeFeatures, createDummyFeatures: arg ‘exclude’ was replaced by ‘cols’
  • normalizeFeatures is now S3 and can be called also on data.frames
  • SMOTEWrapper: fix a bug where “sw.nn” was not correctly passed down
  • fixed a bug that caused hyperparameters to be not passed on correctly in the ModelMultiplexer in some cases
  • fix bug with NoFeaturesModel and ModelMultiplexer
  • fix small bug in DownsampleWrapper when trained with weights
  • getNestedTuneResultsOptPathDf: added new arg “trafo”
  • improve documentation for permutation.importance filter and perform slight argument renaming to fix potential name clashes
  • plotPartialDependence can plot classification tasks with more than one interacted features now
  • generateFilterValuesData: added argument ‘more.args’
  • add pretty.names arguments to plots that show learner short names instead of IDs
  • addition of ‘data’ argument to plotPartialDependence which adds the training data to the graph
  • added new arguments “facet.wrap.nrow” and “facet.wrap.ncol” which enable arrangement of facets in rows and columns to plotting functions

functions - new

  • generateHyperParsEffectData, plotHyperParsEffect
  • makeMultilabelClassifierChainsWrapper, makeMultilabelDBRWrapper makeMultilabelNestedStackingWrapper, makeMultilabelStackingWrapper
  • makeConstantClassWrapper
  • generateFunctionalANOVAData

functions - removed

  • getParamSet generic (now in ParamHelpers package)

functions - renamed

  • generatePartialPrediction to generatePartialDependence
  • plotPartialPrediction to plotPartialDependence
  • plotPartialPredictionGGVIS to plotPartialDependenceGGVIS

learners - general

  • fixed weight handling and weight tag for some learners
  • remove unnecessary linear.output parameter for classif.neuralnet
  • remove unsupported KSVM parameter value stringdot
  • fix some bartMachine compatibility issues
  • classif.ranger, regr.ranger and surv.ranger: now respect unordered factors by default
  • clean up randomForestSRC and randomForestSRCSyn learners
  • the “penalized” learner were restructured and improved (params were added), also see below.
  • add stability.nugget parameter for “”
  • classif.blackboost, regr.blackboost: made sure that arg “stump” is passed on correctly
  • fixed parameter values for WEKA learners IBk, J48, PART, EM, SimpleKMeans, XMeans
  • classif.glmboost, regr.glmboost: add parameters stopintern and trace

learners - new

  • classif.C50
  • classif.gausspr
  • classif.penalized.fusedlasso
  • classif.penalized.lasso
  • classif.penalized.ridge
  • classif.h2o.deeplearning
  • classif.h2o.gbm
  • classif.h2o.glm
  • classif.h2o.randomForest
  • classif.rrf
  • regr.penalized.fusedlasso
  • regr.gausspr
  • regr.glm
  • regr.GPfit
  • regr.h2o.deeplearning
  • regr.h2o.gbm
  • regr.h2o.glm
  • regr.h2o.randomForest
  • regr.rrf
  • surv.penalized.fusedlasso
  • surv.penalized.lasso
  • surv.penalized.ridge
  • cluster.kkmeans
  • multilabel.randomforestSRC

learners - removed

  • surv.optimCoxBoostPenalty
  • surv.penalized (split up, see new learners above)

measures - general

  • updated gmean measure and unit test, added reference to formula of gmean
  • makeCostMeasure: removed arg “task”, names of cost matrix are checked on measure calculation

measures - new

  • multiclass.brier
  • brier.scaled
  • logloss
  • multilabel.subset01, multilabel.f1, multilabel.acc, multilabel.ppv, multilabel.tpr
  • multiclass.au1p, multiclass.au1u, multiclass.aunp, multiclass.aunu

measures - renamed

  • multiclass.auc to multiclass.au1u
  • hamloss to multilabel.hamloss