TPOT

TPOT, short for Tree-based Pipeline Optimization Tool, is an open-source AutoML library for Python that automates the design of machine learning pipelines. It constructs pipelines from scikit-learn components, including preprocessing steps, feature transformers, and estimators, and searches the space of possible pipelines using genetic programming.

During optimization, each candidate pipeline is evaluated using cross-validated performance with a user-selected scoring metric. TPOT uses evolutionary operators such as selection, crossover, and mutation to evolve a population of pipelines across generations. The approach aims to discover high-performing models while exploring a variety of preprocessing and feature engineering strategies.

The best pipeline is returned as a standard scikit-learn Pipeline object and can be used directly for prediction. TPOT can also generate a human-readable Python script that reproduces the discovered workflow.

Usage typically involves selecting TPOTClassifier for classification tasks or TPOTRegressor for regression tasks, creating an instance with parameters such as generations, population_size, offspring_size, mutation_rate, crossover_rate, scoring, cv, and n_jobs, then fitting on training data. After fitting, best_pipeline_ holds the top model, and its steps can be inspected or exported.

TPOT is designed to work with common data formats (NumPy arrays or pandas DataFrames) and integrates tightly with scikit-learn pipelines. It is distributed as a Python package and is available via PyPI and GitHub; it is intended for data scientists seeking automated baseline models or exploratory pipeline design in AutoML workflows.