tensorflow_example.py

import numpy as np
import tensorflow as tf


# Build the model architecture.
input1 = tf.placeholder(tf.float32, [None, 5], name='input1')
input2 = tf.placeholder(tf.float32, [None, 10], name='input2')
conc = tf.concat([input1, input2], axis=-1)
weights = tf.get_variable(
    'weights',
    initializer=np.random.normal(0, 0.01, size=(15, 1)).astype(np.float32))
bias = tf.get_variable('bias', initializer=tf.zeros(1))
logits = tf.reshape(tf.matmul(conc, weights) + bias, [-1])
output = tf.nn.sigmoid(logits)

# Set the learning objective.
target = tf.placeholder(tf.float32, [None], name='target')
loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(
    labels=target, logits=logits))

# Set the gradient descent update rule.
learning_rate = tf.placeholder(tf.float32, [], name='learning_rate')
update_op = tf.train.GradientDescentOptimizer(learning_rate).minimize(loss)

# Generate some random data.
x1 = np.random.uniform(-1, 1, size=(10, 5)).astype(np.float32)
x2 = np.random.uniform(-1, 1, size=(10, 10)).astype(np.float32)
y = np.random.choice([0, 1], size=10).astype(np.float32)

# Start the tensorflow session and initialize the model parameters.
with tf.Session() as session:
    session.run(tf.global_variables_initializer())
    # Update the model for 10 iterations.
    for i in xrange(10):
        train_error = session.run(
            [loss, update_op],
            feed_dict={input1: x1, input2: x2, target: y, learning_rate: 0.01})
        print 'Training Error: %.4f' % train_error[0]