Example notebooks

The package includes six curated notebooks and their input datasets. After installation, locate them without assuming a site-packages path:

from importlib.resources import files

examples = files("mistic.examples")
print(examples)

Run a notebook from a writable copy rather than editing the installed package. The repository versions can also be opened directly on GitHub.

Classification

Forward selection

RBF classification, combined-rank forward selection, blind evaluation, and boundary-counterfactual and integrated-gradient plots.

Backward selection

Backward elimination across rank weights, with the same evaluation and boundary-counterfactual and attribution workflow for a direct comparison.

Probability workflow

Probability-enabled SVC tuning, Brier-aware scoring, calibrated blind probabilities, and probability integrated gradients.

Additional visualizations

Pair plots, three-dimensional views, clustered attributions, and local feature-attribution relationships.

Regression

Boston housing regression

Target and feature transformations, linear/RBF/polynomial SVR comparison, and regression feature analysis.

One-class classification

Breast-cancer novelty detection

Compares both choices of inlier class under controlled splits, with blind metrics, perturbation summaries, and integrated gradients.

Data files

wdbc.data supports the breast-cancer examples and boston-housing_train.csv supports the regression example. The datasets are included for reproducibility; review their provenance and suitability before using them beyond these demonstrations.

Synthetic model comparison

The Synthetic benchmark against scikit-learn models tutorial provides a reproducible known-signal comparison between MISTIC, a plain RBF SVC, linear-SVM RFE followed by an RBF SVC, random forests, and histogram gradient-boosted trees.