Linear Regression with Tensorflow

linear_regression.py

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

# setup dummy data
n = 1000
x = np.random.normal(0, 0.55, n)
y = 0.1*x + 0.3 + np.random.normal(0, 0.03, n)  #want to find a=0.1 and b=0.3
plt.plot(x, y)
plt.show()

def linear_regression(x, y, alpha=0.5, steps=8):
    # initialize a: random slope between -1 and 1
    a = tf.Variable(tf.random_uniform([1], -1, 1))
    # initialize b: 0
    b = tf.Variable(tf.zeros([1]))
    
    yh = a*x + b
    loss = tf.reduce_mean(tf.square(yh - y))
    
    train = tf.train.GradientDescentOptimizer(alpha).minimize(loss)
    
    sess = tf.Session()
    sess.run(tf.global_variables_initializer())

    for i in range(steps):
        sess.run(train)
        #print(i, sess.run(a), sess.run(b))
        ah = sess.run(a)
        bh = sess.run(b)
        
    return ah[0], bh[0]

a, b = linear_regression(x, y)

print("a: %f" % a)
print("b: %f" % b)