ml-for-research

What problem does it solve?

This Skill helps you build trustworthy machine learning models for research by avoiding data leakage, estimating performance fairly with nested cross-validation, and producing explainability and reporting artifacts that meet publication expectations.

Core Features & Use Cases

• Reproducible research pipelines: scikit-learn Pipeline + ColumnTransformer that correctly impute, scale, one-hot encode, and prevent train/test leakage during cross-validation.
• Unbiased evaluation: nested cross-validation (outer repeated stratified folds + inner hyperparameter tuning) to reduce optimistic bias in reported metrics.
• Strong baselines and tuned models: FLAML AutoML with a time budget plus Optuna-based hyperparameter search options, with comparison to simple baselines (e.g., logistic regression / dummy).
• Explainability and transparency: SHAP TreeExplainer outputs (e.g., summary/waterfall/dependence) to interpret feature contributions.
• Calibrated probabilities and model cards: calibration curves using CalibratedClassifierCV, and structured model card generation for transparent method/limitations disclosure.

Quick Start

Use the ml-for-research skill to run a nested cross-validation experiment with a scikit-learn preprocessing pipeline, then generate SHAP summaries and a calibrated model-card report for your dataset.