KerasTuner Documentation

KerasTuner allows you to automate hyperparameter optimization. The workflow involves:

1. Defining a model building function: Create a function that accepts an hp argument. Use this hp object to define your search space (e.g., using hp.Choice).
2. Initializing a Tuner: Choose a search algorithm (such as RandomSearch, BayesianOptimization, or Hyperband) and specify the objective (the metric to optimize) and max_trials (the number of hyperparameter combinations to test).
3. Running the search: Call tuner.search() with your training data and hyperparameters.
4. Retrieving the best model: Use tuner.get_best_models() to access the top-performing model.

import keras_tuner
from tensorflow import keras

# 1. Define the model building function with hyperparameter definitions
def build_model(hp):
  model = keras.Sequential()
  model.add(keras.layers.Dense(
      hp.Choice('units', [8, 16, 32]),
      activation='relu'))
  model.add(keras.layers.Dense(1, activation='relu'))
  model.compile(loss='mse')
  return model

# 2. Initialize the tuner
tuner = keras_tuner.RandomSearch(
    build_model,
    objective='val_loss',
    max_trials=5)

# 3. Start the search
tuner.search(x_train, y_train, epochs=5, validation_data=(x_val, y_val))

# 4. Get the best model
best_model = tuner.get_best_models()[0]