Glass-box options
| Model | What you can show | Accuracy (tabular) |
|---|---|---|
| Logistic regression | coefficient per feature (odds ratio per unit / per SD) | good with engineered features |
| Points scorecard | integer points per bin; total -> probability | close to logistic |
| GAM / EBM | a shape curve per feature, a heatmap per interaction pair | often close to GBMs |
| Shallow decision tree / rule list | IF-THEN rules | lower |
| Monotone GBM + SHAP reason codes | per-decision top reasons, guaranteed directions | high |
EBM (InterpretML)
from interpret.glassbox import ExplainableBoostingClassifier
ebm = ExplainableBoostingClassifier(interactions=10, random_state=0)
ebm.fit(X_train, y_train) # handles strings and NaN
glob = ebm.explain_global() # shape functions; interpret.show(glob) in notebooks
local = ebm.explain_local(X_test[:5], y_test[:5]) # exact additive contributions per rowShape functions can be edited (e.g. to remove an implausible bump) and monotonicity enforced after fitting
(ebm.monotonize(term, increasing=True) in recent versions).
Points scorecard (credit-style)
- Bin each feature (quantiles or optimal binning, e.g.
optbinning), enforcing monotone event rates where required. - Replace bins by weight of evidence, fit logistic regression.
- Scale:
points = offset + factor * (coef * WoE)withfactor = PDO / ln(2)(points to double the odds).
Monotone constraints
| Library | Parameter |
|---|---|
| HistGradientBoosting | monotonic_cst=[1, 0, -1, ...] (or a dict by feature name, >= 1.2) |
| LightGBM | monotone_constraints=[1, 0, -1, ...], monotone_constraints_method="advanced" |
| XGBoost | monotone_constraints="(1,0,-1)" or a dict |
| CatBoost | monotone_constraints=["feature:1", ...] |
| EBM | monotonize() after fit |
Constraints usually cost little accuracy when they match domain knowledge and make models safer under drift.
Reason codes
For adverse-action notices (e.g. credit decisions), report the top features that pushed the score toward the decision. For GBMs: TreeSHAP values per row, mapped to human-readable reasons; for scorecards: the bins with the largest point loss relative to the maximum. Document the method and test stability of the top reasons.
Regulation checklist (high level)
- Model inventory and documentation (model card).
- Prohibited features and proxies reviewed (protected attributes, close proxies such as zip in some contexts).
- Per-group performance and error rates.
- Explanation method validated (stable, faithful).
- Change management and monitoring.
Specific obligations depend on jurisdiction and sector; involve compliance early.
anomaly-detection-model-builder
Build and evaluate anomaly detection for tabular, sensor and transaction data with few or no labels: define the anomaly type (point, contextual, collective), choose detectors (robust z-score/MAD and IQR rules, Mahalanobis with robust covariance, k-NN distance, LOF, Isolation Forest, one-class SVM, PCA reconstruction / autoencoders, forecasting residuals for time series), combine them by rank, explain each alert by its deviating features, set thresholds from an alert budget or the score distribution, and evaluate with precision@k, PR-AUC and recall on known incidents, or detector agreement when no labels exist. Includes pure-Python detectors (robust z, Mahalanobis, kNN, Isolation Forest, ensemble) and an evaluation script.
catboost-model-builder
Build CatBoost models that use its strengths correctly: raw categorical columns as cat_features (ordered target statistics, no manual encoding), text_features, Pool construction with NaN-safe categories, ordered boosting and has_time for temporal data, symmetric (oblivious) trees, overfitting detector with eval_set and use_best_model, losses for classification, regression (Quantile, Poisson, Tweedie) and ranking (YetiRank), class weights, monotone constraints, feature importance (PredictionValuesChange, LossFunctionChange) and SHAP values, GPU training and model export. Plans cat/text/numeric columns from a CSV and generates a runnable training script, and checks parameter sets for the conflicts CatBoost rejects (iterations vs n_estimators, Bayesian bootstrap with subsample, multiple class-weight options, depth > 16, use_best_model without eval_set).
categorical-encoding-selector
Choose and implement the right encoding for every categorical column given the model family and the data: one-hot (with infrequent-level grouping and handle_unknown), ordinal with an explicit order, cross-fitted target encoding (scikit-learn TargetEncoder, CatBoost ordered statistics, category_encoders inside CV), frequency/count encoding, feature hashing, native categorical support in LightGBM, HistGradientBoosting, XGBoost and CatBoost, prefix levels for codes (zip3, NAICS-2), and embeddings. Measures cardinality, rare-level share, unseen-level rate on a time-ordered holdout and target signal per column, emits the ColumnTransformer, and quantifies target-encoding leakage (in-sample vs out-of-fold vs holdout AUC).
clustering-algorithm-selector
Choose the right clustering algorithm and check that clustering makes sense at all: cluster tendency (Hopkins statistic with a correlated-features caveat, nearest-neighbour isolation), then k-means / MiniBatchKMeans, Gaussian mixtures with BIC, agglomerative (Ward, average, complete, single) with dendrograms, DBSCAN, HDBSCAN, OPTICS, spectral clustering, BIRCH, mean shift, k-prototypes and k-modes for mixed or categorical data, and Gower distance, matched to data size, expected shapes, noise, varying density, known k, soft memberships, hierarchy and streaming. Covers preprocessing (scaling, encoding, dimensionality, distance metrics), validation (internal indices, bootstrap stability, external labels, business profiling) and assigning new points.
collaborative-filtering-implementer
Implement collaborative filtering recommenders that beat popularity: item-based kNN with cosine, Jaccard or asymmetric similarity, shrinkage, neighbourhood size, positive-only or mean-centred rating signals and 'because you liked X' explanations; implicit ALS with confidence weighting, BPR pairwise ranking and the closed-form EASE model in plain NumPy, tuned on a temporal validation split by recall@k and NDCG@k and refit on all data; plus user-kNN, explicit matrix factorisation, the implicit/LightFM libraries, hyperparameter ranges, cold-start fallbacks, scaling with sparse matrices and approximate nearest neighbours. Produces top-k lists in the format of the recommendation evaluator.
counterfactual-explanation-generator
Generate counterfactual explanations and algorithmic recourse: the smallest realistic change to a row's features that changes the model's decision. Solves exact minimal-cost recourse for linear and logistic models (MAD-scaled L1 cost, immutable features, increase/decrease directions, bounds, integer and binary features, diverse alternatives, features every counterfactual must change, and a no-recourse finding when immutable features decide), generates DiCE counterfactuals for any scikit-learn model with constraints turned into permitted ranges and every result re-predicted, and validates counterfactuals from any method for constraint violations, implausible values, off-manifold rows, broken derived-feature relations and failure to flip the prediction.