lr-tuning

What problem does it solve?

This Skill helps you improve LR scorecard performance by diagnosing overfitting or underfitting and then tuning both WoE binning settings and LR regularization to achieve better out-of-time (OOT) stability.

Core Features & Use Cases

• WoE + LR Joint Tuning: Uses Bayesian optimization (Optuna TPE) to jointly search WoE binning parameters (max_n_bins, iv_threshold) and LR parameters (C, regularization).
• Diagnosis-Driven Constraints: Adjusts the search space based on model status (overfit / underfit / well-fit) using Train vs Val gaps while keeping OOT strictly for reporting.
• Interactive and AUTO Modes: Supports single-round interactive tuning (pause for feedback) and multi-round AUTO tuning until convergence.

Quick Start

Run lr-tuning after you have a baseline LR scorecard model trained with lr-modeling, and ask the agent to automatically tune it on your dataset by calling the LR tuning skill with your data_path and target.