tpot

Why we included this project

Anyone who has burned an afternoon trying the same few classifiers and preprocessing tricks on a structured dataset will see the point of TPOT quickly: it does that searching for you. Instead of making you fix the feature scaling, the model, and its hyperparameters before you start, it treats the whole sklearn-style pipeline as one thing to optimize and evolves candidate pipelines with genetic programming until it finds a good one. The recent rewrite added genetic feature selection, multi-objective optimization, and a more flexible way to define the search space, so you can point the search at your data rather than accept whatever a black box returns. Analysts and data scientists get a credible, well-documented AutoML baseline to compare against hand-built models, and the dask-backed parallelism means the search can spread across cores. If you want a reproducible starting point for structured data before investing in bespoke engineering, this is worth running.