Learning Data Mining with Python - Second Edition

Creating the neural network

We will now build the convolutional neural network. I have performed some tinkering and found a layout that works well, but feel free to experiment with more layers (or fewer), layers of different types and different sizes. Smaller networks train faster, but larger networks can achieve better results.

First, we create the layers of our neural network:

from keras.layers import Dense, Flatten, Convolution2D, MaxPooling2Dconv1 = Convolution2D(32, 3, 3, input_shape=(d, h, w), activation='relu')pool1 = MaxPooling2D()conv2 = Convolution2D(64, 2, 2, activation='relu')pool2 = MaxPooling2D()conv3 = Convolution2D(128, 2, 2, activation='relu')pool3 = MaxPooling2D()flatten = Flatten()hidden4 = Dense(500, activation='relu') ...