tpot

Uses genetic programming to explore thousands of pipeline configurations, often uncovering high-performing models that manual methods might miss, as demonstrated in its biomedical research applications.

Supports custom configuration spaces and a wide range of scikit-learn operators, allowing tailored searches for specific data types and problem domains, per the README's emphasis on modularity.

Can optimize for multiple goals simultaneously, such as accuracy and model complexity, providing balanced solutions without manual tuning, a key feature highlighted in the documentation.

Leverages Dask for distributed computing, enabling faster optimization on large datasets by parallelizing pipeline evaluations, as noted in the README's best practices.