import keras
from keras.datasets import mnist
import matplotlib.pyplot as plt

from keras.models import Sequential
from keras.layers import Dense, Dropout, Activation, Flatten
from keras.layers import Conv2D, MaxPooling2D


(x_train, y_train), (x_test, y_test) = mnist.load_data()

img_rows = x_train.shape[1]
img_cols = x_train.shape[2]
channels = 1 # since images are in greyscale

print('Training data oryginal shape', x_train.shape)
print('Testing data oryginal shape', x_test.shape)

# reshape data to be in form (num,row,col,chanel)
x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, channels)
x_test = x_test.reshape(x_test.shape[0], img_rows, img_cols, channels)


target_train = keras.utils.to_categorical(y_train, 10)

target_test = keras.utils.to_categorical(y_test, 10)


data_train = x_train.astype('float32')
data_test = x_test.astype('float32')


convNN = Sequential()

convNN.add(Conv2D(5, (5, 5), strides = (1, 1), padding='same', input_shape=(28, 28, 1), \
                  activation='relu'))
convNN.add(Conv2D(5, (3, 3), strides = (3, 3)))
convNN.add(Activation('relu'))
convNN.add(MaxPooling2D(pool_size=(3, 3)))

convNN.add(Flatten())
convNN.add(Dense(50))
convNN.add(Activation('relu'))
convNN.add(Dense(10))
convNN.add(Activation('softmax'))

convNN.summary()


#hyberparameters parameters
batch_size = 500
epochs = 10

# select and initiate optimizer
opt_sgd = keras.optimizers.sgd(lr=0.05)


convNN.compile(loss='categorical_crossentropy', optimizer=opt_sgd, metrics=['accuracy'])


run_hist_sgd = convNN.fit(data_train, target_train, batch_size=batch_size, epochs=epochs,
                          validation_data=(data_test, target_test), shuffle=True,
                          verbose=1) #verbose = 0: no verbose; 1: progress barr; 2: summary afer epch

#model accuracy
scores = convNN.evaluate(data_test, target_test, verbose=True)
print('Test accuracy:', scores[1])