The hardware and bandwidth for this mirror is donated by dogado GmbH, the Webhosting and Full Service-Cloud Provider. Check out our Wordpress Tutorial.
If you wish to report a bug, or if you are interested in having us mirror your free-software or open-source project, please feel free to contact us at mirror[@]dogado.de.
PipeOpImputeConstant’s constant hyperparameter
since it was incompatible with other defaults and would lead to not
recommended usage (creating an empty level).paradox versions
pre-1.0.0.empty_level_control argument to
PipeOpImpute allowing control over edge cases for
factor/ordered columns.empty_level_control to
"param" for PipeOpImputeOOR and to
"always" for PipeOpImputeConstant.PipeOps that take NULL as input
during training now automatically perform training during
prediction.PipeOpImputeConstant, PipeOpImputeMode,
PipeOpImputeOOR, and PipeOpImputeLearner can
now handle factor or ordered features with
zero levels.PipeOpImputeConstant now gives a more informative error
message if check_levels is TRUE and a new
level would be created through imputation.PipeOpImputeOOR now imputes
".MISSING" for factor/ordered
features with only NAs instead of sampling from the
feature’s levels.PipeOpImputeLearner no longer adds
"factor" or "ordered" levels for these feature
types arbitrarily and instead updates levels correctly in certain
edge-cases.PipeOps training and
prediction.PipeOp’s error message
wrapper: now correctly says “This happened in …”.PipeOps / Graphs.use_parallel of PipeOpVtreat.bbotk::OptimizerBatchNLoptr in
LearnerClassifAvg / LearnerRegrAvg’s internal
optimize_weights_learneravg function.preproc() for easier
training of or prediction with PipeOps or
Graphs.PipeOpVtreat, PipeOpEncodeImpact, and
PipeOpEncodeLmer now accept the more precise
TaskSupervised instead of Task as input for
training and prediction.task_type of
the input and output channels of PipeOps that inherit from
PipeOpTaskPreproc and set a non-default
task_type.PipeOpEncodeLmer, PipeOpADAS,
PipeOpBLSmote, PipeOpSmote, and
PipeOpSmoteNC no longer throw an error in case of empty
target levels during training.PipeOpClassBalancing now handles unseen target
levels by ignoring them during upsampling instead of producing
NAs.no_collapse_above_absolute for
PipeOpCollapseFactors /
po("collapse_factors").PipeOpCollapseFactors now correctly collapses
levels of ordered factors.LearnerClassifAvg and LearnerRegrAvg
hyperparameters get the "required" tag.use_groups (default TRUE)
for PipeOpSubsampling to respect grouping (changed default
behaviour for grouped data)new_role_direct for
PipeOpColRoles / po("colroles") to change
column roles by role instead of by column.po() / pos() /
ppl() / ppls() now make suggestions for
entries in both mlr_pipeops as well as
mlr_graphs when an object by the given name could not be
found in the respective dictionary.PipeOpDecode / po("decode") to
reverse one-hot or treatment encoding.feature and something else no
longer lose the other column role during training or predicting of
PipeOps inheriting from
PipeOpTaskPreproc.PipeOpBLSmote deterministic.PipeOpFilter.PipeOpEncodePLQuantiles and
PipeOpEncodePLTree that implement piecewise linear encoding
with two different binning methods.R6 release.PipeOpNMF and PipeOpLearnerPICVPlus.PipeOpTargetMutate and
PipeOpTargetTrafoScaleRange no longer drop unseen factor
levels of features or targets during train and predict.PipeOpTargetMutate.mlr3PipeOpTomek / po("tomek") and
PipeOpNearmiss / po("nearmiss")PipeOpLearnerPICVPlus / po("learner_pi_cvplus")PipeOpLearnerQuantiles /
po(learner_quantiles)GraphLearner has new active bindings/methods as
shortcuts for active bindings/methods of the underlying
Graph: $pipeops, $edges,
$pipeops_param_set, and
$pipeops_param_set_values as well as $ids()
and $plot().PipeOpRowApply /
po("rowapply")PipeOp IDs now explicitly forbidden.Graph$tran() / Graph$predict()
with single_input = FALSE now correctly handles
PipeOps with multiple inputs.GraphLearner$base_learner() now works with
PipeOpBranch, and is generally more robust.GraphLearner now supports $importance,
$selected_features(), $oob_error(), and
$loglik(). These are computed from the underlying
Learner.GraphLearner$impute_selected_features option added:
$selected_features() is reported even if the underlying
base learner does not report it; in this case, the full feature set as
seen by that learner is returned.GraphLearner$predict_type handling more robust
now.PipeOpThreshold and PipeOpTuneThreshold
now have the $predict_type "prob". They can be
set to "response", in which case the probability
predictions are discarded, potentially saving memory.PipeOpImputeOOR now retains the
.MISSING level in factors during prediction that were
imputed during training, but had no missing values during
prediction.as_data_table(po()) now works even when some
PipeOps can not be constructed. For these
PipeOps, NA is reported in most columns.mlr3 release.PipeOpADAS /
po("adas"), PipeOpBLSmote /
po("blsmote") and PipeOpSmoteNC /
po("smotenc")bbotk release.GraphLearnerppl("convert_types").inst/. These are
considered experimental and unstable.PipeOpFeatureUnion used in
ppl("robustify") and ppl("stacking").pipeline_bagging() gets the replace
argument (old behaviour FALSE by default).$add_pipeop() method got an argument
clone (old behaviour TRUE by default).PipeOpFeatureUnion in some rare cases dropped
variables called "x".ppl("robustify") pipelines.PipeOpTuneThreshold was not overloading the
correct .train and .predict functions.$hash and $phash for
GraphLearner and all PipeOps. This could break
users that inherit from PipeOp and make use of
$hash in the future (but is ultimately in their
interest!).phash of GraphLearner now
considers content of Graph, not only IDs.po(), pos() can now construct
PipeOps with ID postfix _<number> to
avoid ID clashes.GraphLearner now has method
$base_learner() that returns the underlying
Learner, if it can be found by a simple heuristic.PipeOpHistBin operation.PipeOpPCA documentation of
center default.$label active binding, setting it to the
help()-page title by default.$help() function for all PipeOps as well as
Graph, GraphLearner and all Learners.GraphLearner can be created without cloning
Graph (for internal use).predict.Graph throws helpful error when it cannot
create a fitting Task.PipeOpLearner packages slot is set to the
Learner’s packages.PipeOp train() and
predict() report correct channel name when output has wrong
type.%>>!% that modifies Graphs
in-place.chain_graphs(),
concat_graphs(), Graph$chain() as alternatives
for %>>% and %>>!%.pos() and ppls() which create
lists of PipeOps/Graphs and can be seen as “plural” forms of
po() and ppl().po() S3-method for PipeOp class that
clones a PipeOp object and optionally modifies its attributes.Graph$add_pipeop() now clones the PipeOp being
added.graph_model in GraphLearner
class, which gets the trained Graph.as_learner() S3-method for PipeOp class
that wraps a PipeOp in a Graph and turns that
into a Learner.PipeOpHistBin: renamed bins Param to
breaksPipeOpImputeHist: fix handling of integer features
spanning the entire represented integer rangePipeOpImputeOOR: fix handling of integer features
spanning the entire represented integer rangePipeOpProxy: Avoid unnecessary clonePipeOpScale: Performance improvementbbotk version.mlr_graphs: pipeline_stackingmlr3 version.PipeOpFilter gets additional
filter.permuted hyperparameter.add_edge of Graphs work with
Multiplicities.GraphLearner hash depend on
id.LearnerAvg.mlr3 version.bbotk 0.3.0as.data.table(mlr_pipeops) work with
paradox 0.6PipeOpColApply: now allows for an applicator function
with multiple columns as a return value; also inherits from
PipeOpTaskPreprocSimple nowPipeOpMissInd now also allows for setting type =
integerPipeOpNMF: now exposes all parameters previously in
.optionsmlr_graphs:
pipeline_bagging now uses multiplicities
internallypipeline_robustify determines the type of newly
created columns when using PipeOpMissIndPipeOpFeatureUnion: Fixed a minor bug when checking for
duplicatesexpect_valid_pipeop_param_setGraphLearnerGraphLearner allows custom idmlr3 0.6NULL input channels accept any kind of inputprint() method of Graphs now also allows for printing a
DOT representation on the consolestate of PipeOps is now reset to NULL when
training failsas_learner.PipeOpLearnerClassifAvg, LearnerRegrAvg use
bbotk nowppl_robustify detects whether a learner can
handle factorsPipeOpTextVectorizer can now return an “integer
sequence representation”.PipeOpNMFPipeOpColRolesPipeOpVtreatmlr_graphs:
pipeline_baggingpipeline_branchpipeline_greplicatepipeline_robustifypipeline_targettrafopipeline_ovrPipeOpOVRSplit, PipeOpOVRUnitePipeOpReplicatePipeOpMultiplicityExply,
PipeOpMultiplicityImplyPipeOpTargetTrafo, PipeOpTargetInvertPipeOpTargetMutatePipeOpTargetTrafoScaleRangePipeOpProxyPipeOpDateFeaturesPipeOpImputeConstantPipeOpImputeLearnerPipeOpModePipeOpRandomResponsePipeOpRenameColumnsPipeOpTextVectorizerPipeOpThresholdPipeOpImputeNewlvl –> PipeOpImputeOOR
(with additional functionality for continuous values)PipeOpFeatureUnion: Bugfix: avoid silently overwriting
features when names clashPipeOpHistBin: Bugfix: handle test set data out of
training set rangePipeOpLearnerCV: Allow returning trainingset prediction
during train()PipeOpMutate: Allow referencing newly created
columnsPipeOpScale: Allow robust scalingPipeOpLearner, PipeOpLearnerCV:
learner_models for access to learner with model slotselector_missingselector_cardinality_greater_than%>>%PipeOpTaskPreproc now has feature_types
slotPipeOpTaskPreproc(Simple) internal API changed: use
.train_task(), .predict_task(),
.train_dt(), .predict_dt(),
.select_cols(), .get_state(),
.transform(), .get_state_dt(),
.transform_dt() instead of the old methods without dot
prefix.train(),
.predict() instead of train_internal(),
predict_internal()Graph new method update_ids()Graph methods train(single_input = FALSE)
and predict(single_input = FALSE) now handle vararg
channels correctly.greplicate(); use
pipeline_greplicate / ppl("greplicate")
instead.po() now automatically converts Selector
to PipeOpSelectpo() prints available mlr_pipeops
dictionary contentmlr_graphs dictionary of useful Graphs, with short form
accessor ppl()mlr3 version 0.4.0stringsAsFactors option default change in 3.6 ->
4.0)predict() generic for GraphsaveRDS(), serialize()
etc.mlr3 version 0.1.5 (handling of character
columns changed)PipeOpEncodeImpactPipeOpEncode: handle NAsThese binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.
Health stats visible at Monitor.