ml2.py

import numpy as np 
import pandas as pd 

df = pd.read_csv('dataset.txt', header=None)
df.columns = ['size', 'price']
length = len(df['size'])
for i in df.columns:
	df[i] = df[i] / max(df[i])

def linear_regression(theta_0, theta_1, dataset):
	final_error = 0
	for i in range(length):
		final_error += ((theta_0 + theta_1 * df['size'].iloc[i]) - df['price'].iloc[i]) ** 2
	
	return final_error / length

def gradient_descent(theta_0, theta_1, alpha):
	error_sum_0 = 0
	error_sum_1 = 0
	for i in range(length):
		error_sum_0 +=  (theta_0 + theta_1 * df['size'].iloc[i]) - df['price'].iloc[i]
		error_sum_1 += ((theta_0 + theta_1 * df['size'].iloc[i]) - df['price'].iloc[i]) * df['size'].iloc[i]

	result_theta_0 = theta_0 - (alpha * (1/length * error_sum_0))
	result_theta_1 = theta_1 - (alpha * (1/length * error_sum_1))
	return [result_theta_0, result_theta_1]

def gradient_descent_runner(theta_0, theta_1, alpha, num_iterations):
	new_theta_0 = theta_0
	new_theta_1 = theta_1
	for i in range(num_iterations):
		new_theta_0, new_theta_1 = gradient_descent(theta_0, theta_1, alpha)

	return [new_theta_0, new_theta_1]


print('cost before gradient descent = {}'.format(linear_regression(0, 0, df)))

init_theta_0 = 0
init_theta_1 = 0
alpha = 0.0001
num_iterations = 1000

theta_0, theta_1 = gradient_descent_runner(init_theta_0, init_theta_0, alpha, num_iterations)
print('cost after gradient descent = {}'.format(linear_regression(theta_0, theta_1, df)))
predicted_price = (df.iloc[7]['size'] * theta_1) + theta_0
real_price = df.iloc[7]['price']

print('real price = {} ; predicted price = {}'.format(real_price, predicted_price))
print('difference in prediciton = {}'.format(predicted_price - real_price))