What problem does it solve? Researchers working with neuroimaging-derived tabular data need reproducible classical statistical inference and machine learning without data leakage from preprocessing steps applied outside cross-validation folds. ## Core Features & Use Cases - Classification and Regression: Train logistic regression, Ridge, Elastic Net, SVM/SVR, and XGBoost models with scikit-learn pipelines that keep imputation, scaling, and feature selection inside each training fold. - Statistical Inference: Compute Cohen's d effect sizes, robust formula-based OLS/GLM, and linear mixed-effects models for dose, time, and treatment analyses via statsmodels. - Site-Aware Validation: Use group-aware cross-validation so recording sites are never split between training and test folds. - Use Case: Given a CSV of subject-level ROI features and diagnosis labels from multiple acquisition sites, train a logistic regression classifier with site-aware 5-fold cross-validation and obtain AUROC metrics, fold assignments, and a run manifest. ## Quick Start Run the training script on your features CSV with a target column, model type such as logistic or ridge, and an output directory to get cross-validated metrics and predictions.
