API reference#

Decision trees#

neural_trees.SoftDecisionTree([depth, ...])

Soft Decision Tree Classifier (sklearn-compatible).

neural_trees.SoftDecisionTreeRegressor([...])

Soft decision tree for regression, scikit-learn compatible.

neural_trees.NumpySoftTree(depth, weights, ...)

Torch-free copy of a fitted SoftDecisionTree or SoftDecisionTreeRegressor (kind).

neural_trees.HardRegressionTree(weights, ...)

A trained SoftDecisionTreeRegressor with its gates read as hard decisions and one value per leaf, in target units.

neural_trees.MultivariateDecisionTree([...])

Multivariate Decision Tree Classifier (sklearn-compatible).

neural_trees.OmnivariateDecisionTree([...])

Omnivariate Decision Tree Classifier (sklearn-compatible).

neural_trees.HardDecisionTree(weights, ...)

A trained Soft Decision Tree with its gates read as hard decisions.

Mixtures of experts#

neural_trees.HierarchicalMixtureOfExperts([...])

Hierarchical Mixture of Experts with Dropout Regularization.

neural_trees.HardRoutedExperts(gate_weights, ...)

A trained mixture of experts with its gates read as hard routing decisions.

Constructive and classical models#

neural_trees.GALNetwork([initial_hidden, ...])

GAL (Grow and Learn) Constructive Neural Network.

neural_trees.WeightedKNN([k, weight_power, ...])

Distance-Weighted K-Nearest Neighbors Classifier.

neural_trees.NaiveBayesClassifier([...])

Naive Bayes Classifier with selectable likelihood.

Distillation#

neural_trees.DistilledTree([teacher, ...])

One soft tree trained on what a strong model learned: fit trains the teacher on (X, y), then the student tree on the teacher's class probabilities on the same rows, on synthetic rows near them (synthetic) and on unlabelled rows if given (mixed with the true labels by alpha).

neural_trees.DistilledForest([teacher, ...])

A few soft trees trained on what a strong model learned: the same teacher and transfer rows as DistilledTree, then n_trees students, each on its own bootstrap sample of those rows.

neural_trees.distill(teacher, X, *[, ...])

Fit a soft tree to the probabilities of a fitted classifier.

neural_trees.fidelity(teacher, student, X)

Share of rows of X where student predicts the class teacher predicts.

neural_trees.Distilled(student, fidelity, ...)

The student and how closely it follows the teacher.

neural_trees.distillation.munge(X[, passes, ...])

Synthetic rows near the rows of X, for a teacher to label (MUNGE, Bucila, Caruana and Niculescu-Mizil, "Model compression", KDD 2006).

Statistical tests#

neural_trees.combined_5x2cv_f_test(clf_A, ...)

Alpaydın's Combined 5×2 Cross-Validation F Test.

neural_trees.mcnemar_test(y_true, y_pred_A, ...)

McNemar's Test for comparing two classifiers on the same test set.

neural_trees.paired_t_test(clf_A, clf_B, X, y)

Paired t-test for classifier comparison.

Explanations#

neural_trees.Explanation(predicted_class, ...)

neural_trees.GateStep(node, went, ...)

One gate on the dominant path.

neural_trees.Counterfactual(feature, index, ...)