basic linear regression

linear_regression.py

import matplotlib.pyplot as plt
from sklearn.datasets import make_regression #dataset
import random


X,Y = make_regression(n_features=1, noise=10, n_samples=1000)

plt.scatter(X,Y,s=5 , label = "training")
plt.xlabel('Feature - X')
plt.ylabel('Target - Y')

const = [random.random(), random.random()] #const[0] is m and const[1] is b (assuming y = mx + b)

def hypothesis(x , m , c):
	return (m*x) + c

def cost(m , c):
	error = 0
	i = 0
	while i < len(X):
		error += (hypothesis(X[i], m , c) - Y[i])**2
		i+=1
	return error

def SGD(x , y , learning_rate = 0.001):
	dCdm = 2*(hypothesis(x , const[0] , const[1]) - y)*x #dC/dm where C is cost function, found this out by chain rule
	dCdc = 2*(hypothesis(x , const[0] , const[1]) - y)   #Similarly dC/db
	const[0] -= learning_rate * dCdm
	const[1] -= learning_rate * dCdc


def epoch(j):
	i = 0
	while i < len(X):
		SGD(X[i] , Y[i])
		i+=1
	print("Epoch number "+str(j+1)+" cost equals " + str(cost(const[0], const[1])))

print("Initial Cost = " + str(cost(const[0] , const[1])))


for i in range(100):
	epoch(i+1)

y = []
for x in X:
	y.append(hypothesis(x , const[0] , const[1]))

plt.scatter(X,y,s=5 , label = "prediction")

plt.show()
print(const[0] , const[1])