0.2.0 - 2019-05-27

• PyPI
• GitHub

compose

• compose.Pipeline now has a debug_one.
• compose.Discard and compose.Select now take variadic inputs, which means you don't have to provide a list of features to exclude/include.

datasets

• Added datasets.fetch_bikes

feature_extraction

• Classes that inherit from feature_extraction.VectorizerMixin can now directly be passed str instances instead of dict instances.
• feature_extraction.Agg and feature_extraction.TargetAgg can now aggregate on multiple attributes.

metrics

• Added RollingAccuracy
• Added RollingCrossEntropy
• Added RollingF1
• Added RollingLogLoss
• Added RollingMacroF1
• Added RollingMacroPrecision
• Added RollingMacroRecall
• Added RollingMAE
• Added RollingMicroF1
• Added RollingMicroPrecision
• Added RollingMicroRecall
• Added RollingMSE
• Added RollingPrecision
• Added RollingRecall
• Added RollingRMSE
• Added RollingRMSLE
• Added RollingSMAPE

model_selection

• Added model_selection.online_qa_score.

proba

The dist module has been renamed to proba and is now public, for the moment it contains a single distribution called proba.Gaussian.

naive_bayes

• Added naive_bayes.BernoulliNB.
• Added naive_bayes.ComplementNB.

optim

• Added optim.AdaBound.

tree

• Added tree.DecisionTreeClassifier.
• Removed tree.MondrianTreeClassifier and tree.MondrianTreeRegressor because their performance wasn't good enough.

stats

• Added stats.AutoCorrelation.
• Added stats.EWVar.
• Rename stats.Variance to stats.Var and stats.RollingVariance to stats.RollingVar.

stream

• Added stream.simulate_qa.

utils

• Added utils.SDFT.
• Added utils.Skyline.
• Renamed the window_size parameter to size in utils.Window and utils.SortedWindow.