lr-tuning

To improve LR scorecard out-of-time (OOT) stability, you can use Optuna TPE Bayesian optimization to jointly tune WoE binning parameters and logistic regression regularization, diagnosing overfitting or underfitting via Train vs Val gaps.

The best way to tune WoE binning and logistic regression hyperparameters is using Optuna TPE Bayesian optimization to search over max_n_bins, iv_threshold, C, and regularization to maximize AUC and KS metrics.

Yes, you need a baseline LR scorecard model trained with lr-modeling before tuning, as the tuning skill requires an existing baseline to diagnose and optimize for better out-of-time performance.

Diagnosis-driven constrained search adjusts the hyperparameter search space based on model status by comparing Train vs Val gaps, keeping OOT data strictly for reporting to ensure reliable credit risk modeling evaluation.

Yes, you can automate hyperparameter tuning by using the AUTO mode, which performs multi-round Optuna TPE optimization until convergence, or use the interactive mode for single-round tuning with feedback.

Poor OOT performance usually indicates logistic regression overfitting or underfitting, which can be diagnosed by analyzing Train vs Val gaps and corrected by jointly tuning WoE binning and regularization parameters.