neural_architecture_search.py

import numpy as np
from keras.models import Sequential
from keras.layers import (
    Dense,
    Conv2D,
    MaxPooling2D,
    Flatten,
    Activation,
    Dropout,
)
from keras.datasets import cifar10
from keras.utils import to_categorical

from hyperactive import Hyperactive

(X_train, y_train), (X_test, y_test) = cifar10.load_data()

y_train = to_categorical(y_train, 10)
y_test = to_categorical(y_test, 10)

# to make the example quick
X_train = X_train[0:1000]
y_train = y_train[0:1000]

X_test = X_test[0:1000]
y_test = y_test[0:1000]


def conv1(nn):
    nn.add(Conv2D(32, (3, 3)))
    nn.add(Activation("relu"))
    nn.add(MaxPooling2D(pool_size=(2, 2)))
    return nn


def conv2(nn):
    nn.add(Conv2D(32, (3, 3)))
    nn.add(Activation("relu"))
    return nn


def conv3(nn):
    return nn


def cnn(opt):
    nn = Sequential()
    nn.add(
        Conv2D(
            opt["filters.0"],
            (3, 3),
            padding="same",
            input_shape=X_train.shape[1:],
        )
    )
    nn.add(Activation("relu"))
    nn.add(Conv2D(opt["filters.0"], (3, 3)))
    nn.add(Activation("relu"))
    nn.add(MaxPooling2D(pool_size=(2, 2)))
    nn.add(Dropout(0.25))

    nn.add(Conv2D(opt["filters.0"], (3, 3), padding="same"))
    nn.add(Activation("relu"))
    nn = opt["conv_layer.0"](nn)
    nn.add(Dropout(0.25))

    nn.add(Flatten())
    nn.add(Dense(opt["neurons.0"]))
    nn.add(Activation("relu"))
    nn.add(Dropout(0.5))
    nn.add(Dense(10))
    nn.add(Activation("softmax"))

    nn.compile(
        optimizer="adam", loss="categorical_crossentropy", metrics=["accuracy"]
    )
    nn.fit(X_train, y_train, epochs=5, batch_size=256)

    _, score = nn.evaluate(x=X_test, y=y_test)

    return score


search_space = {
    "conv_layer.0": [conv1, conv2, conv3],
    "filters.0": [16, 32, 64, 128],
    "neurons.0": list(range(100, 1000, 100)),
}


hyper = Hyperactive()
hyper.add_search(cnn, search_space, n_iter=5)
hyper.run()