Reason codes
Reason codes are the top features pushing a decision in the unfavourable direction, translated into approved wording.
def reason_codes(sv_row, names, wording, k=4, direction=+1):
"""Top-k features pushing the score towards the adverse outcome (direction=+1: higher score = adverse)."""
contrib = sorted(((direction * v, n) for v, n in zip(sv_row, names) if direction * v > 0), reverse=True)
return [wording.get(n, n) for _, n in contrib[:k]]- Map grouped features (all one-hot levels, all aggregates of a source) to a single code.
- Keep a reviewed wording table. Features without approved wording should not be used as reasons, and arguably not as features.
- Use a fixed, documented background (e.g. a sample of approved applicants) so that codes mean "relative to the typical approved case".
- Adverse-action notices (credit decisions in the US under ECOA/Regulation B; similar duties under GDPR Art. 22 and the EU AI Act for high-risk systems) require the principal reasons. Have compliance review the method. SHAP is commonly used, but the wording and the background are what get scrutinised.
Latency and storage
| Pattern | When |
|---|---|
| Compute SHAP at scoring time (TreeExplainer path-dependent) | per-request explanations, small trees, < ~10 ms budget |
| Compute in batch after scoring | dashboards, monitoring, outreach lists |
| Store top-k only (feature, value, SHAP) | audit trails; full vectors only for sampled rows |
| Precompute for the decision-relevant segment | e.g. only declined or flagged cases |
LightGBM and XGBoost compute SHAP natively and quickly: model.predict(X, pred_contrib=True) (LightGBM) and
booster.predict(DMatrix(X), pred_contribs=True) (XGBoost) return SHAP values plus a bias column without the
shap package. This is the path-dependent variant.
Monitoring explanations
- Track mean |SHAP| per feature per week. A feature whose share jumps signals drift or a pipeline bug before performance metrics move (labels arrive late).
- Track the base value. It should change only with a model or background change.
- Compare SHAP distributions between training and production with the same population-stability checks used for features.
- After a retrain, diff the global ranking and directions against the previous model. Unexplained flips need review before release.
Governance checklist
- explainer, perturbation mode, background (source, size, date) and output space documented
- additivity checked in the pipeline
- correlated features grouped for reporting
- reason-code wording reviewed; sensitive attributes and their proxies are never reasons
- explanations stored with the model version
- the model card (model-card-writer) links the global explanation
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).
classification-model-selector
Choose the right classification algorithm for tabular, text or mixed data and prove the choice with a fair bake-off: logistic regression (L2, L1, elastic net), linear SVM, naive Bayes, k-NN, RBF SVM, decision trees, random forest, extra trees, HistGradientBoosting, LightGBM, XGBoost, CatBoost, explainable boosting machines (EBM/GAM), points scorecards and MLPs, matched to data size, width, categorical and high-cardinality share, missingness, class imbalance, text, and constraints such as interpretability, monotonic constraints, calibrated probabilities, latency budgets and tiny deployable models. Ranks candidates with starting hyperparameters and cautions, then generates a bake-off script running every candidate on identical CV folds with per-fold metrics, fit time and prediction latency.
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.