Skip to content

Unreleased

base

  • The base.BinaryClassifier and base.MultiClassifier have been merge into base.Classifier. The 'binary_only' tag is now used to indicate whether or not a classifier support multi-class classification or not.

compose

  • Fixed some bugs related to mini-batching in compose.Pipeline.

datasets

  • Added datasets.SolarFlare, which is a small multi-output regression dataset.

decomposition

  • decomposition.LDA now takes as input word counts instead of raw text.

expert

  • Created this new module, which will regroup methods that perform expert learning, which boils down to managing multiple models.
  • Moved ensemble.StackingBinaryClassifier to expert.StackingClassifier.
  • Moved model_selection.SuccessiveHalvingClassifier to expert.SuccessiveHalvingClassifier.
  • Moved model_selection.SuccessiveHalvingRegressor to expert.SuccessiveHalvingRegressor.
  • Moved ensemble.HedgeRegressor to ensemble.EWARegressor.

evaluate

  • Created this new module, which will contains methods for evaluating models.

feature_extraction

  • Moved preprocessing.PolynomialExtender to feature_extraction.PolynomialExtender.
  • Moved preprocessing.RBFSampler to feature_extraction.RBFSampler.

linear_model

  • Added linear_model.Perceptron, which is implemented as a special case of logistic regression.

model_selection

  • Deleted this module.

multiclass

  • multiclass.OneVsRestClassifier now supports mini-batching.

optim

  • Removed optim.MiniBatcher.
  • Implemented optim.Averager, which allows doing averaged stochastic gradient descent.
  • Removed optim.Perceptron.

utils

  • Moved model_selection.expand_param_grid to utils.expand_param_grid.