guide.importance

guide.importance(estimator, X, y, *, sample_weight=None, permutations=300, max_depth=4, random_state=0)

GUIDE variable importance scores (Loh and Zhou, 2021).

Scores come from unpruned trees of depth max_depth grown with the configuration of estimator, which must be an unfitted or fitted GuideRegressor (constant leaves, squared-error loss) or GuideClassifier (default priors and costs). The trees have constant nodes, so an LDA classifier is scored like a constant one. The estimator itself is not modified. permutations response permutations estimate each score’s null mean for the bias adjustment; they run on estimator.n_jobs threads.

Returns:

dict

The report written by guide-rs importance: scores (per feature, in decreasing order of adjusted score, with raw_score, null_mean, adjusted_score, normalized_score and group), thresholds, permutations, seed and warnings.