keras_tuner_randomsearch_mnist_hyperparameter_optimization.py

1import keras
2from keras import layers
3import keras_tuner
4
5def build_model(hp):
6    model = keras.Sequential()
7    model.add(layers.Flatten())
8    model.add(
9        layers.Dense(
10            # Define the hyperparameter
11            units=hp.Int("units", min_value=32, max_value=512, step=32),
12            activation="relu",
13        )
14    )
15    model.add(layers.Dense(10, activation="softmax"))
16    model.compile(
17        optimizer="adam",
18        loss="sparse_categorical_crossentropy",
19        metrics=["accuracy"],
20    )
21    return model
22
23# Initialize the tuner
24tuner = keras_tuner.RandomSearch(
25    hypermodel=build_model,
26    objective="val_accuracy",
27    max_trials=3,
28    executions_per_trial=2,
29    overwrite=True,
30    directory="my_dir",
31    project_name="helloworld",
32)
33
34# Load data
35(x_train, y_train), (x_test, y_test) = keras.datasets.mnist.load_data()
36
37# Normalize pixel values
38x_train = x_train.astype("float32") / 255
39x_test = x_test.astype("float32") / 255
40
41# Start the search
42tuner.search(x_train, y_train, epochs=2, validation_data=(x_test, y_test))
43
44# Get the best model
45best_model = tuner.get_best_models(num_models=1)[0]