#!/usr/bin/env python3
"""Recommend a scaler / transform for every numeric column, for a given model family.
Per column it measures: range, mean/std, median/IQR, skew, share of robust outliers
(|x - median| / (1.4826 * MAD) > 3.5), zero share, negatives, whether values are bounded in
[0, 1], and integer-ness. It also reports the spread of standard deviations across columns
(why unscaled distance-based models are dominated by the widest column).
Model families:
tree random forest, gradient boosting, decision trees: no scaling needed (monotone-invariant)
linear linear/logistic regression with regularisation, SVM, PCA, k-NN, k-means, neural nets:
scale everything; choose the transform per column shape
nn like linear, but prefers bounded / Gaussian-like inputs (QuantileTransformer for heavy tails)
sparse sparse matrices (TF-IDF, one-hot heavy): MaxAbsScaler (keeps zeros)
Per-column choices: none, StandardScaler, RobustScaler, log1p -> StandardScaler,
PowerTransformer(yeo-johnson), PowerTransformer(box-cox), QuantileTransformer(normal), MinMaxScaler,
MaxAbsScaler, passthrough (binary 0/1).
Usage:
python scaling_plan.py data.csv --model linear --drop customer_id --target churned
python scaling_plan.py data.csv --model tree
python scaling_plan.py data.csv --model linear --target price --code > prep.py
Exit codes: 0 plan printed, 2 usage or read error.
"""
import argparse
import csv
import json
import math
import re
import sys
from collections import defaultdict
MISSING = {"", "na", "n/a", "nan", "null", "none", "?"}
def num(v):
try:
x = float(v)
return x if math.isfinite(x) else None
except ValueError:
return None
def q(xs, p):
pos = (len(xs) - 1) * p
lo, hi = math.floor(pos), math.ceil(pos)
return xs[lo] + (xs[hi] - xs[lo]) * (pos - lo)
def stats(xs):
xs = sorted(xs)
n = len(xs)
mean = sum(xs) / n
sd = math.sqrt(sum((x - mean) ** 2 for x in xs) / n)
med = q(xs, 0.5)
mad = q(sorted(abs(x - med) for x in xs), 0.5) * 1.4826
skew = 0.0 if sd == 0 else sum(((x - mean) / sd) ** 3 for x in xs) / n
out = sum(1 for x in xs if mad and abs(x - med) / mad > 3.5) / n
return {"n": n, "min": xs[0], "max": xs[-1], "mean": mean, "std": sd, "median": med,
"iqr": q(xs, 0.75) - q(xs, 0.25), "skew": skew, "outlier_share": out,
"zero_share": sum(1 for x in xs if x == 0) / n, "negatives": xs[0] < 0,
"integer": all(x.is_integer() for x in xs), "distinct": len(set(xs))}
def choose(s, model):
if s["distinct"] <= 2:
return "passthrough", "binary indicator: leave as 0/1"
if model == "tree":
return "none", "trees split on thresholds; any monotone rescaling gives the same model"
if model == "sparse":
return "MaxAbsScaler", "keeps sparsity (zeros stay zero), scales to [-1, 1]"
heavy = abs(s["skew"]) > 1 or s["outlier_share"] > 0.05
if 0 <= s["min"] and s["max"] <= 1 and not heavy:
return ("none" if model == "nn" else "StandardScaler"), "already bounded in [0, 1]" + (
"" if model == "nn" else "; standardise so its scale matches other features")
if heavy:
if s["min"] >= 0 and s["skew"] > 1:
if model == "nn" and s["outlier_share"] > 0.05:
return "QuantileTransformer(normal)", "heavy right tail with outliers: rank-based transform to N(0,1)"
if s["zero_share"] > 0:
return "log1p -> StandardScaler", f"right-skewed (skew {s['skew']:.1f}) with zeros: log1p then standardise"
return "PowerTransformer(box-cox)", f"strictly positive, right-skewed (skew {s['skew']:.1f})"
if s["outlier_share"] > 0.05 and abs(s["skew"]) <= 1:
return "RobustScaler", f"{s['outlier_share']:.0%} robust outliers: centre on median, scale by IQR"
return "PowerTransformer(yeo-johnson)", f"skewed (skew {s['skew']:.1f}) with negatives: Yeo-Johnson"
return "StandardScaler", "roughly symmetric, few outliers: zero mean, unit variance"
CODE = {
"StandardScaler": "StandardScaler()",
"RobustScaler": "RobustScaler(quantile_range=(25.0, 75.0))",
"log1p -> StandardScaler": 'make_pipeline(FunctionTransformer(np.log1p, inverse_func=np.expm1, '
'feature_names_out="one-to-one"), StandardScaler())',
"PowerTransformer(box-cox)": 'PowerTransformer(method="box-cox")',
"PowerTransformer(yeo-johnson)": 'PowerTransformer(method="yeo-johnson")',
"QuantileTransformer(normal)": 'QuantileTransformer(output_distribution="normal", n_quantiles=1000, subsample=100_000)',
"MaxAbsScaler": "MaxAbsScaler()",
"MinMaxScaler": "MinMaxScaler()",
"passthrough": '"passthrough"',
"none": '"passthrough"',
}
def main():
ap = argparse.ArgumentParser(description="Recommend per-column scaling for a model family.", epilog=__doc__,
formatter_class=argparse.RawDescriptionHelpFormatter)
ap.add_argument("csv")
ap.add_argument("--model", choices=["tree", "linear", "nn", "sparse"], default="linear")
ap.add_argument("--target", help="excluded from scaling (scale a regression target separately)")
ap.add_argument("--drop", nargs="*", default=[])
ap.add_argument("--code", action="store_true")
ap.add_argument("--json", action="store_true")
a = ap.parse_args()
try:
with (sys.stdin if a.csv == "-" else open(a.csv, newline="", encoding="utf-8-sig")) as fh:
reader = csv.DictReader(fh)
header = reader.fieldnames or []
rows = list(reader)
except OSError as e:
print(f"error: {e}", file=sys.stderr)
return 2
plan = []
for c in header:
if c == a.target or c in a.drop:
continue
vals = [(r.get(c) or "").strip() for r in rows]
present = [v for v in vals if v.lower() not in MISSING]
xs = [num(v) for v in present]
if not present or any(x is None for x in xs):
continue
s = stats(xs)
s["missing"] = 1 - len(present) / len(vals)
choice, why = choose(s, a.model)
plan.append({"column": c, "transform": choice, "why": why, **{k: (round(v, 4) if isinstance(v, float) else v)
for k, v in s.items()}})
if not plan:
print("error: no numeric columns found", file=sys.stderr)
return 2
stds = [p["std"] for p in plan if p["std"] > 0 and p["distinct"] > 2]
ratio = max(stds) / min(stds) if stds else 1.0
if a.code:
groups = defaultdict(list)
for p in plan:
groups[p["transform"]].append(p["column"])
print("import numpy as np\nfrom sklearn.compose import ColumnTransformer\nfrom sklearn.impute import SimpleImputer\n"
"from sklearn.pipeline import make_pipeline\nfrom sklearn.preprocessing import (FunctionTransformer, MaxAbsScaler, "
"MinMaxScaler, PowerTransformer,\n QuantileTransformer, RobustScaler, "
"StandardScaler)\n\nnumeric_steps = [")
for t, cols in groups.items():
name = re.sub(r"\W+", "_", t.lower()).strip("_") # unique per transform, e.g. powertransformer_box_cox
step = CODE[t]
if step != '"passthrough"':
step = f'make_pipeline(SimpleImputer(strategy="median"), {step})'
print(f' ("{name}", {step}, {cols!r}),')
print("]\nprep = ColumnTransformer(numeric_steps, remainder=\"passthrough\")\n"
"# Fit inside a Pipeline: scalers learn their statistics from the training fold only.")
return 0
if a.json:
print(json.dumps({"model": a.model, "std_ratio": round(ratio, 2), "columns": plan}, indent=2))
return 0
print(f"Scaling plan for '{a.model}' models: {len(plan)} numeric column(s); largest/smallest std ratio "
f"{ratio:,.0f}x")
if a.model != "tree" and ratio > 10:
print(" unscaled, distance- and penalty-based models would be dominated by the widest columns")
print(f"{'column':<20} {'min':>10} {'max':>10} {'skew':>6} {'outl%':>6} {'zero%':>6} transform")
for p in plan:
print(f"{p['column']:<20} {p['min']:>10.4g} {p['max']:>10.4g} {p['skew']:>6.2f} {p['outlier_share']:>6.1%} "
f"{p['zero_share']:>6.1%} {p['transform']:<28} {p['why']}")
print("\nFit every scaler inside the Pipeline (training folds only). Run with --code for the ColumnTransformer.")
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).
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.