Compute on out-of-fold or holdout predictions, never on training predictions.
1. Decile calibration
Sort by prediction, cut into 10 groups, compare mean prediction with mean actual:
import pandas as pd
d = pd.DataFrame({"y": y, "p": p})
d["decile"] = pd.qcut(d.p.rank(method="first"), 10, labels=False) + 1
cal = d.groupby("decile").agg(mean_pred=("p", "mean"), mean_true=("y", "mean"))
cal["ratio"] = cal.mean_pred / cal.mean_trueRatios should be ~1 everywhere. A rising curve below 1 at the top and above 1 at the bottom is "regression to the mean" (shrinkage); common with strong regularisation and squared error on skewed targets.
2. Heteroscedasticity
Correlation between |residual| and prediction (or a binned plot of residual SD by prediction decile). Growing spread means relative errors are roughly constant: log targets or deviance losses fit better, and constant-width intervals are wrong.
3. Residuals vs features
For each important feature, plot mean residual by quantile bin:
d["feature_bin"] = pd.qcut(X_valid["sqft"], 10, duplicates="drop")
d.groupby("feature_bin").apply(lambda g: (g.p - g.y).mean())A systematic pattern means the model misses a non-linearity or interaction in that feature (add splines, interactions, or let the GBM split more).
4. Segment bias
MAE and bias per business segment (region, product line, size band). A small overall bias can hide a large
under-prediction for the most valuable segment (the fixture's large segment: −14.8%).
5. Residuals over time
Mean residual by month of the prediction date: drift or a structural break shows up as a trend or step.
6. Worst errors
Read the 20 largest absolute errors row by row: unit errors, duplicated records, outliers that are data issues, or a missing feature that explains them.
7. Influence (linear models)
Cook's distance (statsmodels get_influence().cooks_distance) flags rows that move the fit; inspect them,
do not delete them automatically.
Patterns and actions
| Pattern | Action |
|---|---|
| Under-prediction at the top, over at the bottom | less regularisation, log/deviance loss, features for extremes |
| Spread grows with level | log target / Gamma; relative metrics; adaptive intervals |
| Residual trend vs a feature | non-linear term or interaction |
| One segment biased | segment feature or interaction; separate model |
| Trend over time | retrain cadence, time features, drift investigation |
| A few huge errors | data quality; robust loss |
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.