keras-tuner

Why we included this project

Training a Keras model often comes down to fiddling with hyperparameters like batch size, learning rate, and layer widths. KerasTuner lets you declare those choices inline as you build the model, then runs a search over them with one of several built-in algorithms, including random search, Bayesian optimization, and Hyperband. That define-by-run approach keeps the search space close to your model definition, which is handy when you have nested or complicated architectures. The library is also designed to be extended, so researchers can prototype custom search strategies without building the whole infrastructure from scratch. For anyone who works in Keras and wants to pick hyperparameters more deliberately than trial and error, this is a practical tool.