XOR net in Tensorflow

XORnet.py

import tensorflow as tf
import numpy as np


tf.set_random_seed(1)

def generate_train_data(batch_size=64):
	indices = np.random.randint(4, size=batch_size)
	XOR_X = np.array([[0, 0], [0, 1], [1, 0], [1, 1]])
	XOR_Y = np.array([[0], [1], [1], [0]])
	return XOR_X[indices], XOR_Y[indices]


input = tf.placeholder(dtype=tf.float32, shape=[None, 2])
output = tf.placeholder(dtype=tf.float32, shape=[None, 1])


net = tf.contrib.layers.fully_connected(input, 3,activation_fn=tf.sigmoid)
net = tf.contrib.layers.fully_connected(net, 3,activation_fn=tf.sigmoid)
net = tf.contrib.layers.fully_connected(net, 1, activation_fn=tf.sigmoid)

loss = tf.losses.log_loss(labels=output, predictions=net)
optimizer = tf.train.AdamOptimizer(0.03).minimize(loss)
init = tf.global_variables_initializer()
sess = tf.InteractiveSession()
sess.run(init)

epochs = 0
loss_val = np.inf
while loss_val > 1e-8:
	epochs +=1
	XOR_X,XOR_Y = generate_train_data(16)
	_, loss_val = sess.run([optimizer, loss], feed_dict={input: XOR_X, output: XOR_Y})
	TEST_X = np.array([[0, 0], [0, 1], [1, 0], [1, 1]])
	TEST_Y = np.array([[0], [1], [1], [0]])
	if epochs % 2000 == 0:
		print({'epoch': epochs, 'loss': loss_val})

print("Learnt XOR after %s epochs." % epochs)
sess.close()