import numpy
import matplotlib.pyplot as plt

# load file (housing.data.txt)
def load():
	data = numpy.loadtxt(open("housing.data.txt", "rb"), dtype="float")
	data = data[:, [5,13]]	# get all data for 5th and 13th column as x y
	data.sort()
	return data

# hypothesis function (theta0 + theta1 * x)
def hyp(theta, x):
	return theta[0] + theta[1] * x

# derivative for theta 0
def der0(theta, data):
	sum = 0
	m = data.shape[0]  # get rows data.shape = [r, c]

	for x in range(0, m):
		x_val = data[x][0]
		y_val = data[x][1]
		# xth row, first column (x's column) ; xth row, second column (y's column)
		sum += hyp(theta, x_val) - y_val

	return sum / 2



# derivative for theta 1
def der1(theta, data):
	sum = 0
	m = data.shape[0]  # get rows data.shape = [r, c]

	for x in range(0, m):
		x_val = data[x][0]
		y_val = data[x][1]
		# xth row, first column (x's column) ; xth row, second column (y's column)
		sum += (hyp(theta, x_val) - y_val) * x_val

	return sum / 2

# cost function
def cost(theta, data):
	sum = 0.0
	m = data.shape[0]

	for x in range(0, m):
		x_val = data[x][0]
		y_val = data[x][1]
		sum += (hyp(theta, x_val) - y_val) ** 2

	return sum / (2 * m)

# alpha value
alpha = 0.0001
data = load()
theta = [0, 0]

plt.scatter(data[:, 0], data[:, 1])

# do gradient descent 5000 times
for x in range(0, 500):
	newtheta0 = theta[0] - alpha * der0(theta, data)
	newtheta1 =	theta[1] - alpha * der1(theta, data)
	theta[0] = newtheta0
	theta[1] = newtheta1

print(theta[0])
print(theta[1])

plt.show()