Concordance
- Harrell's C: among comparable pairs (the shorter time had an event), the share where the subject with the shorter time had the higher risk score. Ties in risk count half. 0.5 is random and 1 is perfect. It depends on the censoring distribution and becomes optimistic under heavy censoring.
- Uno's C: inverse-probability-of-censoring weighted and truncated at tau. Use it to compare models
(
sksurv.metrics.concordance_index_ipcw).
Concordance only measures ranking. A model can rank perfectly and still be badly calibrated.
Time-dependent AUC
The cumulative/dynamic AUC at time t separates subjects with an event by t from those still event-free at t
(sksurv.metrics.cumulative_dynamic_auc). Report it at the decision horizons: 30/90/365 days for churn
campaigns, the warranty period for failures.
Brier score and integrated Brier score
BS(t) = mean over subjects of (1{T_i > t} - S_hat(t | x_i))^2, IPCW-weighted for censoring. The integrated
Brier score (IBS) averages it over a time range. Benchmarks: the Kaplan-Meier curve (no covariates) gives
the "null" IBS. A useful model is clearly below it, and values around 0.25 mean no information when S is
near 0.5.
Calibration
At a decision time t, group the test subjects by predicted S(t) (quintiles or deciles) and compare the mean prediction with the Kaplan-Meier estimate in each group. The generated script reports the largest gap. Penalised Cox and AFT models are often calibrated better than forests.
RMST
Restricted mean survival time up to tau is the area under S(t) on [0, tau]: the average event-free time within the horizon. The RMST difference between groups ("premium customers stay 155 days longer on average within the first 700 days") is valid without proportional hazards, which makes it the preferred summary when curves cross.
Protocol
- Hold out a test set stratified on the event, or out-of-time (later cohorts) to mimic deployment.
- Evaluation times must lie inside the test follow-up (IPCW metrics need it). The generated script restricts them automatically.
- Compare against the Kaplan-Meier null model and a simple Cox model before trusting complex ones.
- Report uncertainty: bootstrap the test set for C and IBS when models are close.
Reporting template
| Model | Harrell's C | Uno's C | AUC@90 | AUC@365 | IBS | Max calibration gap @180 |
|---|
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.