Python Deep Learning - Second Edition

Classifying handwritten digits with a convolutional network

In the third chapter, we introduced a simple neural network to classify digits using Keras and we got 96% accuracy. We could improve this with some tricks (more hidden neurons, for example), but let's try with a simple CNN instead:

1. First, we'll do the imports. We'll also set the random seed:

# for reproducibilityfrom numpy.random import seedseed(1)from tensorflow import set_random_seedset_random_seed(1)

2. Then, we'll do the imports, including convolutional and max pooling layers, as shown in the following example:

from keras.datasets import mnistfrom keras.models import Sequentialfrom keras.layers import Dense, Activationfrom keras.layers import Convolution2D, MaxPooling2Dfrom ...