0.15.0 - 2023-01-29

active

• Created this module dedicated to online active learning.
• Added api.active.EntropySampler.

base

• Fixed an issue where an estimator that has attribute a pipeline could not be cloned.
• Added a base.DriftAndWarningDetector to clarify the difference between drift detectors that have a warning_detected property and those that don't.
• Added MultiLabelClassifier.
• Added MultiTargetRegressor.
• Added drift.BinaryDriftDetector.
• Added drift.BinaryDriftAndWarningDetector.

conf

• Introduced this new module to perform conformal predictions.
• Added a conf.Interval dataclass to represent predictive intervals.
• Added conf.RegressionJackknife.

datasets

• Removed unnecessary Numpy usage in the synth submodule.
• Changed np.random.RandomState to np.random.default_rng where necessary.

drift

• Added drift.DriftRetrainingClassifier.
• Renamed drift.PeriodicTrigger to drift.DummyDriftDetector to clarify it is a naive baseline.
• Created a binary submodule to organize all drift detectors which only apply to binary inputs.

ensemble

• Added ensemble.ADWINBoostingClassifier.
• Added ensemble.BOLEClassifier.

evaluate

• evaluate.progressive_val_score and evaluate.iter_progressive_val_score will now also produce a report once the last sample has been processed, in addition to every print_every steps.

feature_extraction

• feature_extraction.BagOfWords now outputs a dictionary, and not a collections.Counter.

forest

• Created this new module to host all models based on an ensemble of decision trees.
• Moved ensemble.AdaptiveRandomForestClassifier to forest.ARFClassifier.
• Moved ensemble.AdaptiveRandomForestRegressor to forest.ARFRegressor.
• Added forest.AMFClassifier.
• Added forest.OXTRegressor.

linear_model

• Renamed use_dist to with_dist in linear_model.BayesianLinearRegression's predict_one method.

multiclass

• Added a coding_method method to multiclass.OCC to control how the codes are randomly generated.

multioutput

• Added MultiClassEncoder to convert multi-label tasks into multi-class problems.

preprocessing

• Renamed alpha to fading_factor in preprocessing.AdaptiveStandardScaler.

rules

• Renamed alpha to fading_factor in rules.AMRules.

sketch

• Renamed alpha to fading_factor in sketch.HeavyHitters.

stats

• Renamed alpha to fading_factor in stats.Entropy.
• Renamed alpha to fading_factor in stats.EWMean.
• Renamed alpha to fading_factor in stats.EWVar.

stream

• Upgraded stream.iter_sql to SQLAlchemy 2.0.

tree

• Remove LabelCombinationHoeffdingTreeClassifier. New code should use multioutput.MulticlassEncoder instead.

utils

• Removed artifacts from the merger.