Building the table
-- one row per customer: duration from signup to churn or to the data cut-off
SELECT customer_id,
signup_date,
COALESCE(churn_date, DATE '2026-09-30') AS end_date,
DATE_DIFF(COALESCE(churn_date, DATE '2026-09-30'), signup_date, DAY) AS duration_days,
CASE WHEN churn_date IS NOT NULL AND churn_date <= DATE '2026-09-30' THEN 1 ELSE 0 END AS event,
plan_at_signup, channel_at_signup
FROM customers
WHERE signup_date < DATE '2026-09-30';- Covariates must be known at the origin (here: signup). "Number of support calls" is not, because it grows with time and depends on survival. It needs time-varying handling.
- Durations of 0 (event on day 0) break some models. Use a small positive value (0.5 days) or start at day 1.
Censoring types
| Type | Example | Handling |
|---|---|---|
| Right censoring (administrative) | still a customer at the cut-off | standard |
| Right censoring (loss to follow-up) | moved away, account merged | standard if non-informative |
| Interval censoring | failure found at an inspection between two visits | interval-censored methods (lifelines fit_interval_censoring) |
| Left truncation (delayed entry) | customers acquired before the data history starts | pass entry times (entry= in lifelines) |
Informative censoring
Censoring is informative when its reason is related to the event. Examples:
- customers "censored" when they downgrade, who were about to churn
- patients withdrawn from a study because of side effects
- machines removed for preventive maintenance because they looked worn
Consequences: KM and Cox estimates are biased. Options: treat the censoring reason as a competing event, redefine the event (any downgrade or churn), or use inverse probability of censoring weights.
Checks before modelling
| Check | Why |
|---|---|
| durations >= 0, events in {0, 1} | data errors (the scripts refuse others) |
| events per covariate >= 10 | stable Cox estimates |
| follow-up distribution by group | groups observed for very different lengths compare badly |
| censoring share by group and over calendar time | informative or administrative patterns |
| event times piled at specific values (e.g. day 30, 365) | contract terms; consider discrete-time models |
| covariates measured after the origin | immortal-time bias, leakage |
Long (counting-process) format
For time-varying covariates, use one row per subject per interval:
| id | start | stop | event | plan |
|---|---|---|---|---|
| C1 | 0 | 120 | 0 | basic |
| C1 | 120 | 300 | 1 | premium |
from lifelines import CoxTimeVaryingFitter
CoxTimeVaryingFitter().fit(long_df, id_col="id", start_col="start", stop_col="stop", event_col="event")Immortal-time bias
Classifying subjects by something that happens later ("customers who used feature X") gives them guaranteed event-free time before X. Model X as time-varying, or use a landmark analysis: fix a landmark time, classify by status at the landmark, and analyse survival after it.
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.