mnist_keras_preprocessing.py

# some variables...
image_height = train_digits.shape[1]  
image_width = train_digits.shape[2]
num_channels = 1  # we have grayscale images
# NOTE: image_height == image_width == 28

# re-shape the images data
train_data = np.reshape(train_digits, (train_digits.shape[0], image_height, image_width, num_channels))
test_data = np.reshape(test_digits, (test_digits.shape[0],image_height, image_width, num_channels))

# re-scale the image data to values between (0.0,1.0]
train_data = train_data.astype('float32') / 255.
test_data = test_data.astype('float32') / 255.

# one-hot encode the labels - we have 10 output classes
# so 3 -> [0 0 0 1 0 0 0 0 0 0], 5 -> [0 0 0 0 0 1 0 0 0 0] & so on
from keras.utils import to_categorical
num_classes = 10
train_labels_cat = to_categorical(train_labels,num_classes)
test_labels_cat = to_categorical(test_labels,num_classes)
train_labels_cat.shape, test_labels_cat.shape