Keras Tuner

Keras Tuner is a scalable hyperparameter optimization framework that automates the search for ideal model architectures and training configurations.

This library eliminates manual trial and error by systematically exploring search spaces for TensorFlow and Keras models. Users implement the kt.HyperParameters class to define ranges for specific variables: layer depth, dropout rates, or learning schedules. The engine supports four primary algorithms (BayesianOptimization, Hyperband, RandomSearch, and a Sklearn wrapper) to identify peak performance. By managing execution logs and model checkpointing automatically, it isolates the exact configuration that maximizes validation accuracy while minimizing resource overhead.

https://keras.io/keras_tuner/