Families
| Family | Examples |
|---|---|
| Velocity (entity x window) | count, amount sum and max per card / account / device / IP / email over 5m, 1h, 24h, 7d, 30d |
| Deviation from the entity's own history | amount / mean, z-score, hour-of-day vs the usual hours, new merchant category, new country |
| Ratios of short to long windows | cnt_1h / cnt_30d: sudden bursts relative to normal activity |
| Account / customer | account age, days since a password or email change, KYC level, number of linked cards |
| Device / session | new device, emulator / VPN flags, typing speed, time from login to payment |
| Merchant | MCC, merchant fraud rate (smoothed, lagged), card-not-present, high-risk category |
| Geography | distance from the last transaction / velocity of travel, country mismatch (card vs IP vs shipping) |
| Network / graph | number of cards per device, shared emails / addresses / phones, connected-component size, fraud labels in the neighbourhood (lagged) |
| Unsupervised scores | Isolation Forest, autoencoder reconstruction error, as features |
Point-in-time rules
- Every aggregate uses events strictly before the current event. Include the current event only if the online system will also have it at decision time (for example, a counter incremented before scoring).
- Label-derived features ("merchant fraud rate") must use labels known at the event time. Lag them by the label delay.
- Simultaneous events (same timestamp) do not see each other:
fraud_velocity_features.pyprocesses timestamp batches. - Offline features must equal online features. Test this by recomputing a sample of production-scored events offline and comparing.
Computation
# pandas, leak-free (identical to fraud_velocity_features.py)
df = df.sort_values("timestamp")
roll = (df.set_index("timestamp").groupby("card_id")["amount"]
.rolling("24h", closed="left").agg(["count", "sum"]))Online counters:
- per entity, sliding windows in Redis (sorted sets by timestamp) or a streaming engine (Flink, Kafka Streams), with an exactly-once update after scoring
- approximate counters (HyperLogLog for distinct counts, Count-Min sketches) for high-cardinality keys
- time-decayed counters (exponentially weighted) as a cheaper alternative to exact windows
Encoding high-cardinality IDs
Merchant, device and IP ids: use target encoding with time-respecting out-of-fold estimates and smoothing, or frequency / risk-tier encodings, or graph features instead of raw ids. Never one-hot millions of ids.
Feature hygiene
- Monitor feature distributions weekly (PSI). Fraudsters move to where the model is blind.
- Remove features that only work through label leakage (e.g. "account closed" flags set after fraud is confirmed).
- Cap extreme values, and keep a missing-value indicator: missing device information is itself a signal.
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.