python-tensorflowHow can I use Python and TensorFlow to create an XOR gate?

This can be done by creating a neural network with two input neurons, one output neuron and one hidden layer with two neurons. With this structure, we can create an XOR gate using TensorFlow and Python.

import tensorflow as tf

# Inputs
x = tf.placeholder(tf.float32, shape=[4,2], name='x')

# Outputs
y_ = tf.placeholder(tf.float32, shape=[4,1], name='y_')

# Weights and biases
W_h = tf.Variable(tf.random_uniform([2,2], -1, 1), name="W_h")
b_h = tf.Variable(tf.zeros([2]), name="b_h")

W_o = tf.Variable(tf.random_uniform([2,1], -1, 1), name="W_o")
b_o = tf.Variable(tf.zeros([1]), name="b_o")

# Hidden Layer
h = tf.sigmoid(tf.matmul(x, W_h) + b_h)

# Output Layer
y = tf.sigmoid(tf.matmul(h, W_o) + b_o)

# Cost Function
cost = tf.reduce_mean(( (y_ * tf.log(y)) +
        ((1 - y_) * tf.log(1.0 - y)) ) * -1)

# Optimizer
train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cost)

# Training
sess = tf.Session()
init = tf.global_variables_initializer()
sess.run(init)

# Training Data
XOR_X = [[0,0],[0,1],[1,0],[1,1]]
XOR_Y = [[0],[1],[1],[0]]

# Run
for i in range(100000):
    sess.run(train_step, feed_dict={x: XOR_X, y_: XOR_Y})

# Print
print('Output:')
print(sess.run(y, feed_dict={x: XOR_X}))

Output example

[[0.01991707]
 [0.980087   ]
 [0.97948706]
 [0.01991817]]

Code explanation

1. Importing the TensorFlow library: import tensorflow as tf
2. Creating the placeholders for inputs and outputs: x = tf.placeholder(tf.float32, shape=[4,2], name='x') and y_ = tf.placeholder(tf.float32, shape=[4,1], name='y_')
3. Creating the weights and biases for the hidden and output layers: W_h = tf.Variable(tf.random_uniform([2,2], -1, 1), name="W_h"), b_h = tf.Variable(tf.zeros([2]), name="b_h"), W_o = tf.Variable(tf.random_uniform([2,1], -1, 1), name="W_o") and b_o = tf.Variable(tf.zeros([1]), name="b_o")
4. Computing the output of the hidden layer: h = tf.sigmoid(tf.matmul(x, W_h) + b_h)
5. Computing the output of the output layer: y = tf.sigmoid(tf.matmul(h, W_o) + b_o)
6. Computing the cost function: cost = tf.reduce_mean(( (y_ * tf.log(y)) + ((1 - y_) * tf.log(1.0 - y)) ) * -1)
7. Creating the optimizer: train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cost)
8. Initializing the variables: init = tf.global_variables_initializer()
9. Creating the training data: XOR_X = [[0,0],[0,1],[1,0],[1,1]] and XOR_Y = [[0],[1],[1],[0]]
10. Running the training loop: for i in range(100000): sess.run(train_step, feed_dict={x: XOR_X, y_: XOR_Y})
11. Printing the output: print(sess.run(y, feed_dict={x: XOR_X}))

Helpful links

1. TensorFlow Tutorials
2. Neural Networks and Deep Learning
3. Building Neural Network with TensorFlow

Edit this code on GitHub