Classify every feature
| Class | Examples | Spec |
|---|---|---|
| Immutable | age, birthplace, past events, tenure so far, protected attributes and their proxies | "mutable": false |
| Mutable, one direction | education (up), account age (up), number of late payments in the future window (down) | "direction": "increase" / "decrease" |
| Mutable, bounded | utilisation 0-100%, price >= floor | "min", "max" |
| Discrete | counts, ratings | "integer": true |
| Binary choice | autopay, plan upgrade | "binary": true |
| Costly | income, relocation | "cost_multiplier": 3 |
| Derived | totals, ratios, aggregates | recompute from inputs, or validate with --relation |
Rule of thumb: if the person cannot change it within the decision horizon, it is immutable for recourse. It can still be a legitimate model input.
Cost scale
- MAD (median absolute deviation) is robust to outliers and makes a unit change "typical" (Wachter et al. 2017).
- Use
stdwhen the MAD is 0 (binary or very skewed features).cf_validate.pyfalls back the same way. - Domain costs override statistics: changing income by 10% is much harder than closing one card.
- The percentile shift (cost = change in the feature's CDF; Ustun et al. 2019) is a good alternative: moving from the 50th to the 60th percentile costs the same on every feature.
Derived features and causal structure
Counterfactual search treats features as independent knobs. In reality:
total_charges = tenure x monthly_charges: changing the total alone is meaningless (--relation).- Increasing education raises age (time passes) and can change income.
- Paying down debt changes utilisation and the debt-to-income ratio together.
Options:
- Model only root causes as mutable and recompute derived features before predicting (custom search).
- Mark derived features immutable and let the search change their inputs.
- Validate with relations and reject inconsistent counterfactuals.
- Use causal recourse methods (Karimi et al. 2021) when a structural causal model exists.
Constraint file (shared by all three scripts)
{"features": {
"tenure_months": {"mutable": false, "scale": 14.5},
"support_calls": {"direction": "decrease", "min": 0, "integer": true, "scale": 1.0},
"last_login_days": {"direction": "decrease", "min": 0, "integer": true, "scale": 6.0},
"monthly_charges": {"direction": "decrease", "min": 20, "max": 120, "scale": 14.0, "cost_multiplier": 2.0},
"plan_basic": {"binary": true, "scale": 1.0}
}}linear_counterfactual.py needs intercept and coefficients in the same file. gen_dice_script.py and
cf_validate.py read only the features block.
Common mistakes
| Mistake | Consequence |
|---|---|
| No immutability | "become younger", "change your region" |
| No direction | "increase your past late payments to lower churn" (a sign quirk of the model) |
| No bounds | negative counts, utilisation above 100% |
| Continuous treatment of counts | "2.37 support calls" |
| Ignoring derived features | flips achieved by inconsistent values (seen with DiCE on the churn fixture) |
| One counterfactual only | the single option may be impossible for this person |
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.