#!/usr/bin/env python3
"""Generate a fair classifier bake-off script: every candidate on identical CV folds, same metrics,
per-fold results saved for statistical comparison.
Candidates (--models): logreg, hgb, rf, et, lgbm, xgb, catboost, ebm. Optional libraries (lightgbm,
xgboost, catboost, interpret) are imported lazily; missing ones are skipped with a message.
Each candidate gets the preprocessing its family needs (scaling + one-hot for logreg; ordinal codes
for sklearn trees; pandas category dtype for LightGBM/XGBoost; raw strings for CatBoost/EBM).
The generated script writes bakeoff_folds.csv (model, fold, metric columns, fit_seconds,
predict_ms_per_1k) - feed it to ml-model-comparison-reporter/scripts/compare_models.py - and
prints a summary table.
Usage:
python bakeoff_scaffold.py data.csv --target churned --models logreg hgb lgbm catboost -o bakeoff.py
python bakeoff_scaffold.py data.csv --target churned --group-col customer_id --drop customer_id
python bakeoff_scaffold.py data.csv --target churned --time-col signup_date --folds 5
Exit codes: 0 written, 2 usage or read error.
"""
import argparse
import csv
import math
import os
import sys
from itertools import islice
MISSING = {"", "na", "n/a", "nan", "null", "none", "?"}
KNOWN = ["logreg", "hgb", "rf", "et", "lgbm", "xgb", "catboost", "ebm"]
TEMPLATE = '''"""Classifier bake-off for {target!r}. Generated by bakeoff_scaffold.py.
Every model sees the same folds; results per fold go to bakeoff_folds.csv for paired comparison.
Run: python {script} (pandas, scikit-learn >= 1.4; optional: lightgbm, xgboost, catboost, interpret)
"""
import time
import numpy as np
import pandas as pd
from sklearn.base import clone
from sklearn.compose import ColumnTransformer
from sklearn.ensemble import ExtraTreesClassifier, HistGradientBoostingClassifier, RandomForestClassifier
from sklearn.impute import SimpleImputer
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import average_precision_score, brier_score_loss, log_loss, roc_auc_score
from sklearn.model_selection import {splitter}
from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import OneHotEncoder, OrdinalEncoder, StandardScaler
DATA, TARGET = {data!r}, {target!r}
DROP = {drop!r}
MODELS = {models!r}
SEED = 0
df = pd.read_csv(DATA)
df = df[df[TARGET].notna()]
{order_line}positive = df[TARGET].value_counts().idxmin()
y = (df[TARGET] == positive).astype(int).to_numpy()
{groups_line}X = df.drop(columns=[TARGET] + [c for c in DROP if c in df.columns])
CAT = [c for c in X.columns if not pd.api.types.is_numeric_dtype(X[c])]
NUM = [c for c in X.columns if c not in CAT]
folds = list({cv_expr})
def onehot_scaled():
return ColumnTransformer([
("num", make_pipeline(SimpleImputer(strategy="median", add_indicator=True), StandardScaler()), NUM),
("cat", make_pipeline(SimpleImputer(strategy="most_frequent"),
OneHotEncoder(handle_unknown="infrequent_if_exist", min_frequency=10)), CAT)])
def ordinal(impute):
num = SimpleImputer(strategy="median") if impute else "passthrough"
return ColumnTransformer([
("cat", OrdinalEncoder(handle_unknown="use_encoded_value", unknown_value=-1, encoded_missing_value=-1), CAT),
("num", num, NUM)])
def as_category(frame, cats):
out = frame.copy()
for c in CAT:
out[c] = pd.Categorical(out[c], categories=cats[c])
return out
def make(name):
"""Return (estimator, kind); kind says how to feed data."""
if name == "logreg":
return make_pipeline(onehot_scaled(), LogisticRegression(C=1.0, max_iter=5000)), "sk"
if name == "hgb":
return make_pipeline(ordinal(False), HistGradientBoostingClassifier(
categorical_features=list(range(len(CAT))) or None, learning_rate=0.05, early_stopping=True,
random_state=SEED)), "sk"
if name == "rf":
return make_pipeline(ordinal(True), RandomForestClassifier(n_estimators=500, min_samples_leaf=3, n_jobs=-1,
random_state=SEED)), "sk"
if name == "et":
return make_pipeline(ordinal(True), ExtraTreesClassifier(n_estimators=500, min_samples_leaf=3, n_jobs=-1,
random_state=SEED)), "sk"
if name == "lgbm":
import lightgbm as lgb
return lgb.LGBMClassifier(n_estimators=400, learning_rate=0.05, num_leaves=31, min_child_samples=20,
subsample=0.8, subsample_freq=1, colsample_bytree=0.8, random_state=SEED,
verbose=-1), "category"
if name == "xgb":
import xgboost as xgb
return xgb.XGBClassifier(n_estimators=400, learning_rate=0.05, max_depth=6, subsample=0.8,
colsample_bytree=0.8, tree_method="hist", enable_categorical=True,
random_state=SEED), "category"
if name == "catboost":
from catboost import CatBoostClassifier
return CatBoostClassifier(iterations=600, learning_rate=0.05, depth=6, random_seed=SEED, verbose=0,
cat_features=CAT), "raw"
if name == "ebm":
from interpret.glassbox import ExplainableBoostingClassifier
return ExplainableBoostingClassifier(random_state=SEED), "raw"
raise ValueError(name)
rows = []
for name in MODELS:
try:
proto, kind = make(name)
except ImportError as e:
print(f"skip {{name}}: {{e}}")
continue
for k, (tr, te) in enumerate(folds):
Xtr, Xte = X.iloc[tr], X.iloc[te]
if kind == "category":
cats = {{c: pd.Index(Xtr[c].dropna().unique()) for c in CAT}}
Xtr, Xte = as_category(Xtr, cats), as_category(Xte, cats)
elif kind == "raw":
Xtr, Xte = Xtr.copy(), Xte.copy()
for c in CAT:
Xtr[c], Xte[c] = Xtr[c].astype(str), Xte[c].astype(str)
est = clone(proto) if kind == "sk" else make(name)[0]
t0 = time.perf_counter()
est.fit(Xtr, y[tr])
fit_s = time.perf_counter() - t0
t0 = time.perf_counter()
p = est.predict_proba(Xte)[:, 1]
pred_ms = (time.perf_counter() - t0) / max(1, len(te)) * 1e6
rows.append({{"model": name, "fold": k, "roc_auc": roc_auc_score(y[te], p),
"average_precision": average_precision_score(y[te], p),
"log_loss": log_loss(y[te], np.clip(p, 1e-15, 1 - 1e-15), labels=[0, 1]),
"brier": brier_score_loss(y[te], p), "fit_seconds": fit_s, "predict_ms_per_1k": pred_ms}})
print(f"done {{name}}")
res = pd.DataFrame(rows)
res.to_csv("bakeoff_folds.csv", index=False)
summary = res.groupby("model").agg(["mean", "std"]).drop(columns="fold", level=0)
print(summary.round(4).to_string())
print("\\nPer-fold results in bakeoff_folds.csv; compare with ml-model-comparison-reporter (paired tests).")
'''
def main():
ap = argparse.ArgumentParser(description="Generate a fair classifier bake-off script.", epilog=__doc__,
formatter_class=argparse.RawDescriptionHelpFormatter)
ap.add_argument("csv")
ap.add_argument("--target", required=True)
ap.add_argument("--models", nargs="+", default=["logreg", "hgb", "rf", "lgbm", "catboost"])
ap.add_argument("--drop", nargs="*", default=[])
ap.add_argument("--group-col")
ap.add_argument("--time-col")
ap.add_argument("--folds", type=int, default=5)
ap.add_argument("-o", "--output")
a = ap.parse_args()
bad = [m for m in a.models if m not in KNOWN]
if bad:
print(f"error: unknown model(s) {bad}; choose from {KNOWN}", file=sys.stderr)
return 2
try:
with open(a.csv, newline="", encoding="utf-8-sig") as fh:
reader = csv.DictReader(fh)
header = reader.fieldnames or []
list(islice(reader, 1))
except OSError as e:
print(f"error: {e}", file=sys.stderr)
return 2
for c in (a.target, a.group_col, a.time_col):
if c and c not in header:
print(f"error: column {c!r} not found", file=sys.stderr)
return 2
drop = list(a.drop)
if a.time_col:
splitter = "TimeSeriesSplit"
cv_expr = f"TimeSeriesSplit(n_splits={a.folds}).split(X)"
order_line = f"df = df.sort_values({a.time_col!r}).reset_index(drop=True)\n"
drop.append(a.time_col)
groups_line = ""
elif a.group_col:
splitter = "StratifiedGroupKFold"
cv_expr = f"StratifiedGroupKFold(n_splits={a.folds}, shuffle=True, random_state=SEED).split(X, y, groups)"
order_line = ""
groups_line = f"groups = df[{a.group_col!r}].to_numpy()\n"
if a.group_col not in drop:
drop.append(a.group_col)
else:
splitter = "StratifiedKFold"
cv_expr = f"StratifiedKFold(n_splits={a.folds}, shuffle=True, random_state=SEED).split(X, y)"
order_line = groups_line = ""
text = TEMPLATE.format(target=a.target, script=os.path.basename(a.output or "bakeoff.py"), splitter=splitter,
data=a.csv, drop=drop, models=a.models, order_line=order_line, groups_line=groups_line,
cv_expr=cv_expr)
if a.output:
with open(a.output, "w", encoding="utf-8") as f:
f.write(text)
print(f"wrote {a.output} ({len(a.models)} models, {splitter})", file=sys.stderr)
else:
sys.stdout.write(text)
return 0
if __name__ == "__main__":
sys.exit(main())
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).
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.
counterfactual-explanation-generator
Generate counterfactual explanations and algorithmic recourse: the smallest realistic change to a row's features that changes the model's decision. Solves exact minimal-cost recourse for linear and logistic models (MAD-scaled L1 cost, immutable features, increase/decrease directions, bounds, integer and binary features, diverse alternatives, features every counterfactual must change, and a no-recourse finding when immutable features decide), generates DiCE counterfactuals for any scikit-learn model with constraints turned into permitted ranges and every result re-predicted, and validates counterfactuals from any method for constraint violations, implausible values, off-manifold rows, broken derived-feature relations and failure to flip the prediction.