cclauss / keras-tuner

Hyperparameter tuning for humans

Keras Tuner

An hyperparameter tuner for Keras.

Example (Current API)

Here's how to perform hyperparameter tuning on the MNIST digits dataset, using the current API.

from tensorflow import keras
from tensorflow.keras import layers

import numpy as np

from kerastuner.tuners import GridSearch
from kerastuner.distributions import Range, Choice

(x, y), (val_x, val_y) = keras.datasets.mnist.load_data()
x = x.astype('float32') / 255.
val_x = val_x.astype('float32') / 255.


"""Basic case:
- We define a `build_model` function
- It returns a compiled model
- It uses hyperparameters defined on the fly
"""


def build_model():
    model = keras.Sequential()
    model.add(layers.Flatten(input_shape=(28, 28)))
    for i in range(Range('num_layers', 2, 20)):
        model.add(layers.Dense(units=Range('units_' + str(i), 32, 512, 32),
                               activation='relu'))
    model.add(layers.Dense(10, activation='softmax'))
    model.compile(
        optimizer=keras.optimizers.Adam(
            Choice('learning_rate', [1e-2, 1e-3, 1e-4])),
        loss='sparse_categorical_crossentropy',
        metrics=['accuracy'])
    return model


tuner = GridSearch(
    build_model,
    objective='val_accuracy',
    num_executions=2)

tuner.search(x=x,
             y=y,
             validation_data=(val_x, val_y))

Example (Future API)

Here's how to perform hyperparameter tuning on the MNIST digits dataset.

from tensorflow import keras
from tensorflow.keras import layers

import numpy as np

from kerastuner.tuner import SequentialRandomSearch

(x, y), (val_x, val_y) = keras.datasets.mnist.load_data()
x = x.astype('float32') / 255.
val_x = val_x.astype('float32') / 255.


"""Basic case:
- We define a `build_model` function
- It returns a compiled model
- It uses hyperparameters defined on the fly
"""


def build_model(hp):
    model = keras.Sequential()
    model.add(layers.Flatten(input_shape=(28, 28)))
    for i in range(hp.Range('num_layers', 2, 20)):
        model.add(layers.Dense(units=hp.Range('units_' + str(i), 32, 512, 32),
                               activation='relu'))
    model.add(layers.Dense(10, activation='softmax'))
    model.compile(
        optimizer=keras.optimizers.Adam(
            hp.Choice('learning_rate', [1e-2, 1e-3, 1e-4])),
        loss='sparse_categorical_crossentropy',
        metrics=['accuracy'])
    return model


tuner = SequentialRandomSearch(
    build_model,
    objective='val_accuracy')

tuner.search(trials=2,
             x=x,
             y=y,
             epochs=5,
             validation_data=(val_x, val_y))