from keras.datasets import mnist

(train_images, train_labels), (test_images, test_labels) = mnist.load_data()

# Training images

train_images.shape

# Training labels (classification)

train_labels

# View images from training set

import matplotlib.pyplot as plt

plt.imshow(train_images[1], cmap='gray')

from keras import models
from keras import layers

# Create sequential network model
network = models.Sequential()

# Add a fully connected or dense layer with rectified linear unit activation 
# function
network.add(layers.Dense(512, activation='relu', input_shape=(28 * 28,)))

# Add a fully connected or dense layer which will return an array of 10 
# probability socres (summing to 1)
network.add(layers.Dense(10, activation='softmax'))

network.compile(
  optimizer = 'rmsprop',
    loss    = 'categorical_crossentropy',
    metrics = ['accuracy'],
)

# Flatten training images

train_images = train_images.reshape((60000, 28 * 28))
train_images = train_images.astype('float32') / 255

# Flatten testing images

test_images = test_images.reshape((10000, 28 * 28))
test_images = test_images.astype('float32') / 255

from keras.utils import to_categorical

train_labels = to_categorical(train_labels)
test_labels  = to_categorical(test_labels)



# epochs:     How many times to go through dataset
# batch_size: Size of each batch (how many to process in parallel)


network.fit(train_images, train_labels, epochs=5, batch_size=128)

test_loss, test_acc = network.evaluate(test_images, test_labels)

print('Test Accurancy = {:.2f}%'.format(test_acc*100.0))