Newt

A lightweight Python toolkit for efficient feature analysis and statistical diagnostics in credit risk modeling.

Key Features

• 6 Binning Algorithms: ChiMerge, Decision Tree, K-Means, Equal Frequency, Equal Width, Optimal.
• Monotonic Support: Comprehensive monotonic binning support (ascending, descending, auto-detect).
• Feature Analysis: Robust binning statistics and transformation workflows.
• Feature Selection: Pipeline-style feature selection (IV, PSI, VIF, Stepwise).
• Scorecard Generation: End-to-end scorecard generation and scoring.
• High Performance: Rust-accelerated core paths with component-specific engine defaults (for example, IV/FeatureSelector/Report use auto).

Installation

pip install newt

Quick Start

• User Guide
• API Reference

Recommended workflow: fit Binner, call binner.woe_transform(X), then build the scorecard with Scorecard.from_model(model, binner). For persistence, use LogisticModel.dump/load and Scorecard.dump/load.

Benchmarks

• Performance vs Toad
• PSI Performance
• Stepwise Selection Performance
• ChiMerge Performance