What problem does it solve? Predicting behavioral or clinical outcomes from brain functional connectivity requires careful feature selection that avoids data leakage across cross-validation folds. This Skill implements Connectome Predictive Modeling (CPM) with edge selection repeated independently inside every training fold, producing unbiased classification or regression predictions from subject-level connectomes. ## Core Features & Use Cases - Fold-Local Edge Selection: Selects positive and negative network edges by correlation p-threshold within each training fold only, preventing leakage. - Classification and Regression: Supports both discrete diagnosis prediction (class and probability) and continuous cognitive score prediction from FC matrices or edge vectors. - Full Provenance Outputs: Writes predictions.csv, fold_assignments.csv, metrics.json, checkpoint.joblib models, config.json, and run_manifest.json for reproducible neuroimaging research. - Use Case: Given resting-state fMRI connectomes for 200 subjects and a cognitive score per subject, run the regression workflow to predict held-out scores and obtain per-fold performance metrics. ## Quick Start Run the CPM regression workflow on my connectomes.npz and labels.csv files with a p-threshold of 0.01 and 5-fold cross-validation, saving results to an output directory.
