#!/usr/bin/env python
# coding: utf-8

# ### 선형회귀 모델에 대한 GrandientDescentOptimizer 활용

# - 1개의 값씩 훈련을 시키는 코드

# In[88]:


import tensorflow as tf
import numpy as np

# Our real model of y = 2 * x + 6
def y_real(x):
    return tf.mul(x, 2) + 6

# Our model of y = w * x + b
def y_model(x, w, b):
    return tf.mul(x, w) + b 

# x and y are placeholders for our training data
x = tf.placeholder("float")

# w and b are the variables storing our values. It is initialised with starting "guesses"
w = tf.Variable(1.0, name="w")
b = tf.Variable(1.0, name="b")

# Our error is defined as the square of the differences
error = tf.square(y_real(x) - y_model(x, w, b))

# The Gradient Descent Optimizer does the heavy lifting
train_op = tf.train.GradientDescentOptimizer(0.01).minimize(error)

# Normal TensorFlow - initialize values, create a session and run the model
model = tf.initialize_all_variables()

errors = []
with tf.Session() as session:
    session.run(model)
    for i in range(2000):
        x_value = np.random.rand()
        _, error_value = session.run([train_op, error], feed_dict={x: x_value})
        errors.append(error_value)
        if i % 100 == 0:
            print "Iteration: {0}, w: {1}, b: {2}".format(i, w.eval(), b.eval())
    
    print "Predicted model: {0}x + {1}".format(w.eval(), b.eval())    


# In[89]:


import matplotlib.pyplot as plt
get_ipython().run_line_magic('matplotlib', 'inline')

error_list = [errors[i] for i in range(len(errors))]
num = 0
for error in error_list:
    if error < 0.00001:
        print num
        break
    num += 1    
plt.plot(error_list)
plt.show()
plt.savefig("errors.png")    


# - batch 입력을 적용하여 n개의 값씩 훈련시키는 코드

# In[90]:


# Our real model of y = 2 * x + 6
def y_real(x):
    return tf.mul(x, 2) + 6

# Our model of y = w * x + b
def y_model(x, w, b):
    return tf.mul(x, w) + b 

# Batch
batch_size = 2

# x and y are placeholders for our training data
x = tf.placeholder("float", [None, 1])

# w and b are the variables storing our values. It is initialised with starting "guesses"
w = tf.Variable([1.0], name="w")
b = tf.Variable([1.0], name="b")

# Our error is defined as the square of the differences
error = tf.reduce_sum(tf.square(y_real(x) - y_model(x, w, b)))

# The Gradient Descent Optimizer does the heavy lifting
train_op = tf.train.GradientDescentOptimizer(0.01).minimize(error)

# Normal TensorFlow - initialize values, create a session and run the model
model = tf.initialize_all_variables()

errors = []
with tf.Session() as session:
    session.run(model)
    for i in range(2000):
        x_value = np.random.randn(batch_size, 1)
        _, error_value = session.run([train_op, error], feed_dict={x: x_value})
        errors.append(error_value)
        if i % 100 == 0:
            print "Iteration: {0}, w: {1}, b: {2}".format(i, w.eval(), b.eval())
    
    print "Predicted model: {0}x + {1}".format(w.eval(), b.eval())    


# In[91]:


error_list = [errors[i] for i in range(len(errors))]
num = 0
for error in error_list:
    if error < 0.0001:
        print num
        break
    num += 1  
plt.plot(error_list)
plt.show()
plt.savefig("errors.png")