tf_failsauce

import tensorflow as tf
import numpy as np

def F(x):
  return x * 3.0 + 4.0

def gen_data(n_batch = 10):
  ret_x = []
  ret_y = []
  for _ in range(n_batch):
    x = np.random.random()
    y = F(x)
    ret_x.append(x)
    ret_y.append(y)
  return np.array(ret_x, np.float32), np.array(ret_y, np.float32)

print gen_data()

tf_x = tf.placeholder(tf.float32, [10])
tf_y = tf.placeholder(tf.float32, [10])

theta1 = tf.Variable(tf.truncated_normal([1], stddev=0.1))
theta2 = tf.Variable(tf.truncated_normal([1], stddev=0.1))

pred_y = theta1 * tf_x + theta2
err = tf.reduce_mean(tf.square(pred_y - tf_y))

optimizer = tf.train.GradientDescentOptimizer(0.01)
train = optimizer.minimize(err)


# Launch the graph.
init = tf.initialize_all_variables()
sess = tf.Session()
sess.run(init)

x_data, y_data = gen_data()
print x_data
print x_data.shape
print tf_x.get_shape()

for i in range(1000):
  x_data, y_data = gen_data()
  sess.run(train, feed_dict={tf_x: x_data, tf_y: y_data})
  print sess.run(err)