import tensorflow as tf
import random
from tensorflow.examples.tutorials.mnist import input_data

tf.set_random_seed(777)

mnist = input_data.read_data_sets("./data", one_hot=True)

learning_rate = 0.01
epochs = 15
batch_size = 100

X = tf.placeholder(tf.float32, [None, 784])
Y = tf.placeholder(tf.float32, [None, 10])

W1 = tf.get_variable(name = "W1", initializer= tf.random_normal([784, 256]))
b1 = tf.get_variable(name = "b1", initializer= tf.random_normal([256]))
a1 = tf.nn.relu(tf.matmul(X, W1) + b1) # (?, 256)

W2 = tf.get_variable(name = "W2", initializer= tf.random_normal([256, 256]))
b2 = tf.get_variable(name = "b2", initializer= tf.random_normal([256]))
a2 = tf.nn.relu(tf.matmul(a1, W2) + b2) # (?, 256)

W3 = tf.get_variable(name = "W3", initializer= tf.random_normal([256, 10]))
b3 = tf.get_variable(name = "b3", initializer= tf.random_normal([10]))
hypothesis = tf.matmul(a2, W3) + b3 # (?, 10)

loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(logits=hypothesis, labels=Y))
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(loss)

# For GPU user
# sess = tf.Session(config=tf.ConfigProto(gpu_options=tf.GPUOptions(allow_growth =True)))
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    
    for epoch in range(epochs):
        avg_loss = 0
        total_batches = int(mnist.train.num_examples / batch_size)
    
        for i in range(total_batches):
            batch_x, batch_y = mnist.train.next_batch(batch_size)
            feed_dict = {X:batch_x, Y:batch_y}
            _loss, _ = sess.run([loss, optimizer], feed_dict = feed_dict)
            avg_loss += _loss / total_batches

        print("Epoch : {:d} \t loss : {:.4f}".format(epoch, avg_loss))
        
    print("Learning finished")   
    
    # Test model and check accuracy    
    correct_prediction = tf.equal(tf.argmax(hypothesis, 1), tf.argmax(Y, 1))
    accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
    print("Accuracy : {:.2f}".format(sess.run(accuracy, feed_dict={X:mnist.test.images, Y:mnist.test.labels})))
    
    # Get one and predict it
    r = random.randint(0, mnist.test.num_examples-1)
    print("Sample : {}, Prediction : {}".format(
        sess.run(tf.argmax(mnist.test.labels[r:r+1], 1)),
        sess.run(tf.argmax(hypothesis, 1), feed_dict={X:mnist.test.images[r:r+1]})))