In [50]:

    

import pandas as pd
import tensorflow as tf 
import numpy as np 
import matplotlib.pyplot as plt


    

In [51]:

    

from sklearn.preprocessing import normalize


    

In [52]:

    

df = pd.read_csv('datasets/dataset2.csv')


# Weights and bias for NN
weight = tf.Variable([1.0])
bias = tf.Variable([0.0])


    

In [53]:

    

hours = df['average_montly_hours'].astype(float).values.reshape(1,-1)


    

In [54]:

    

normalized_hours = normalize(df['average_montly_hours'].astype(float).values.reshape(1,-1), norm='l2', axis=1).reshape(-1,1)[:10]


    

In [55]:

    

left = df['left'].astype(np.float32).values


    

In [56]:

    

xs = tf.placeholder(tf.float32)
ys = tf.placeholder(tf.float32)
xlosses = []
losses = []


    

In [57]:

    

output = ( weight * xs ) 
diff =  output - ys
cost = tf.square(diff)

grad = tf.reduce_sum(((weight * xs) - ys ) * xs)
optimized_weight = weight - 0.0001 * grad
reduced_cost = tf.reduce_mean(cost)


    

In [58]:

    

# Start the session
sess = tf.Session()
sess.run(tf.global_variables_initializer())


    

In [20]:

    

for i in range(1000):
    if i%10 == 0:
#         print ''
#         print 'Step: ', i
#         print 'Gradient: ', sess.run([grad1, grad], {xs: normalized_hours, ys: left})
#         print 'Weight: ' , sess.run([weight])[0][0]
#         print 'Bias: ', sess.run([bias])[0][0]
#         print 'Loss: ', sess.run([reduced_cost], {xs: normalized_hours, ys: left})[0]
#         print 'Optimized Weight: ', sess.run([optimized_weight], {xs: normalized_hours, ys: left})[0]
        losses.append(sess.run([reduced_cost], {xs: normalized_hours, ys: left})[0])
    sess.run(tf.assign(weight, sess.run([optimized_weight], {xs: normalized_hours, ys: left})[0]))


    

In [21]:

    

plt.plot(np.arange(100.0), losses)


    

    
Out[21]:





[<matplotlib.lines.Line2D at 0x1154d6e90>]

In [22]:

    

plt.show()


    

In [62]:

    

sess.close()
sess = tf.Session()
sess.run(tf.global_variables_initializer())


    

In [60]:

    

for i in range(100):
    xlosses.append(sess.run([reduced_cost], {xs: hours, ys: left})[0])
    sess.run(tf.assign(weight, sess.run([optimized_weight], {xs: hours, ys: left})[0]))


    

In [61]:

    

plt.plot(np.arange(100.0), xlosses)
plt.show()