Skip to content

Sklearn additions

Patrick Godwin requested to merge sklearn_additions into master

This merge request adds several changes/additions to some sklearn-based classifiers.

In particular, the changes proposed adds:

  • An option for using a cross-validation grid search to tune MLA hyperparameters
  • An implementation of feature_importance()
  • A change to the SVM model so that it doesn't override Classifier._define_model() and make this classifier more in line with the other sklearn-based classifiers.
  • An alternative to CalibrationMap called SklearnCalibrationMap which uses isotonic or Logistic regression (Platt's method) depending on the number of samples given.

It also adds a few changes to idq-report to produce probability timeseries directly from output from idq-timeseries rather than generating them from quivers.

This branch is similar to the other sklearn classifiers branch but removes any issues causing by rebasing that branch against master (which might cause some issues when doing the actual merge). When this branch is merged in, the other branch (mostly identical, besides missing a few changes) will be deleted.

Merge request reports