Auto-PyTorch

Why we included this project

Auto-PyTorch automates the part of deep learning that often takes weeks: architecture search and hyperparameter tuning. It runs Bayesian optimization over network designs and training settings, seeded by a portfolio of configurations that have performed well across many datasets, and it can combine the top candidates into an ensemble. The tool is built for tabular data, both classification and regression, with time series forecasting support as well, so it fits teams that work with tables and panels rather than images or text. You control the search budget by setting wall-clock time and memory limits, and the library manages model candidates, pruning, and disk usage for the ensemble. For teams that want a reproducible route from raw features to a strong deep-learning baseline without deep architectural expertise, this is a practical starting point.