numeric-normalizer

name	numeric-normalizer
description	Normalize and scale numeric columns using min-max scaling, z-score standardization, or robust scaling. Supports saving and applying scaler parameters for consistent train/test normalization.
version	0.2.3

Numeric Normalizer

Skill Summary

Applies numeric normalization/scaling transformations to specified columns in a tabular dataset. Supports min-max scaling (0-1 range), z-score standardization (mean=0, std=1), and robust scaling (median-centered, IQR-scaled). Can save fitted scaler parameters to a JSON file for later re-application to new data (useful for ML train/test consistency).

Scaling Methods

• min-max: x_scaled = (x - min) / (max - min) → range [0, 1]
• z-score: x_scaled = (x - mean) / std → mean 0, std 1
• robust: x_scaled = (x - median) / IQR → robust to outliers
• log: x_scaled = log(x + 1) → for right-skewed distributions
• max-abs: x_scaled = x / max(|x|) → range [-1, 1]

Inputs

• --input / -i: Path to input file (CSV or Parquet) (required)
• --columns / -c: Numeric columns to normalize (comma-separated; default: all numeric)
• --method / -m: Normalization method (default: min-max)
• --output / -o: Output file path
• --save-params: Save scaler parameters to JSON file
• --load-params: Load previously saved parameters (transform-only mode)

Processing Steps

1. Load dataset and identify target columns
2. Compute scaling parameters (min/max, mean/std, or median/IQR)
3. Apply transformation to each target column
4. Optionally save parameters for reproducibility
5. Save normalized dataset

Output

• Normalized dataset (CSV or Parquet)
• scaler_params.json (optional): Fitted parameters per column
• Console: Per-column stats before/after normalization

Implementation

• Command: python3 ./skills/numeric_normalizer/normalize.py -i <file> -o <output> -m <method>
• Dependencies: pandas, numpy, scikit-learn