Hyperparameter Tuning

Hyperparameters are choices made before training, such as layer width and learning rate. A tuning library evaluates several configurations using the validation set and records which choices work best within a fixed search budget.

The PyTorch notebook uses Optuna and the Keras-on-Torch notebook uses KerasTuner. Both search the same compact MNIST CNN space and evaluate the selected model once on the test set.

• Complete PyTorch and Optuna workflow

• KerasTuner alternative using the Torch backend