#!/usr/bin/env python3

import numpy as np
import keras
from keras.models import Sequential
from keras.layers import (Dense, Dropout, Activation, Flatten,
        Conv2D, MaxPooling2D)
from keras.utils import np_utils
from keras.datasets import mnist


# Import MNIST data
(train_samples, train_labels), (test_samples, test_labels) = mnist.load_data()

# Preprocess MNIST images
train_samples = train_samples.reshape(train_samples.shape[0], 28, 28, 1)
test_samples = test_samples.reshape(test_samples.shape[0], 28, 28, 1)
train_samples = train_samples.astype(np.float32)
test_samples = test_samples.astype(np.float32)
train_samples = train_samples / 255
test_samples = test_samples / 255

# Convert labels into array of 10-dim vectors
cat_train_labels = np_utils.to_categorical(train_labels, 10)
cat_test_labels = np_utils.to_categorical(test_labels, 10)

# Build convolutional network
convnet = Sequential()
convnet.add(Conv2D(32, 4, 4, activation='relu', input_shape=(28,28,1)))
convnet.add(MaxPooling2D(pool_size=(2,2), strides=1))
convnet.add(Conv2D(32, 3, 3, activation='relu'))
convnet.add(MaxPooling2D(pool_size=(2,2), strides=1))
convnet.add(Dropout(0.3))
convnet.add(Flatten())
convnet.add(Dense(10, activation='softmax'))

convnet.compile(loss=keras.losses.CategoricalCrossentropy(from_logits=False), optimizer='sgd',
                metrics=['accuracy'])
convnet.summary()

convnet.fit(train_samples, cat_train_labels, batch_size=32, epochs=20,
            verbose=1)
metrics = convnet.evaluate(test_samples, cat_test_labels, verbose=1)
print('{}: {:.2f}'.format(convnet.metrics_names[1], metrics[1]*100))
print(convnet.metrics_names)
predictions = convnet.predict(test_samples)