Keras-02-DenseLayer.py

'''Keras model for MNIST
Based on examples from: 
https://github.com/keras-team/keras/tree/master/examples
'''

from __future__ import print_function

import keras
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense
from keras.optimizers import RMSprop

# The MNIST data is from the numpy archive file mnist.npz 
# The data is shuffled and split between a train and test set,
# consisting of 60,000 training entries and 10,000 test entries
# Data shape (60000, 28, 28), (10000, 28, 28), type uint8
(x_train, y_train), (x_test, y_test) = mnist.load_data()

# The label is a number 0-9, convert to a one-hot vector of size 10
num_classes = 10
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)

# For the Dense layer, flatten each 28x28 image into a vector size 784
x_train = x_train.reshape(60000, 784)
x_test = x_test.reshape(10000, 784)

# Normalize the gray scale value to float value 0.-1.
# First convert the data to float
x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
# Then divide by the max value of uint8:  255
x_train /= 255
x_test /= 255

# Build the model:
model = Sequential()
model.add(Dense(num_classes, activation='softmax', input_shape=(784,)))

model.summary()

model.compile(loss='categorical_crossentropy',
              optimizer=RMSprop(),
              metrics=['accuracy'])

batch_size = 128
epochs = 10
history = model.fit(x_train, y_train,
                    batch_size=batch_size,
                    epochs=epochs,
                    verbose=1,
                    validation_data=(x_test, y_test))
score = model.evaluate(x_test, y_test, verbose=0)
print('Test loss:', score[0])
print('Test accuracy:', score[1])