import numpy as np
# y = wx + b
def cal_err_linear_given_points(b, w, points):
totalErr = 0
for i in range (0, len (points)):
x = points[i, 0]
y = points[i, 1]
totalErr += (y - (w * x + b)) ** 2
return totalErr / float (len (points))
def step_grad(b_current, w_current, points, learnRate):
b_grad = 0
w_grad = 0
N = float (len (points))
for i in range (0, len (points)):
x = points[i, 0]
y = points[1, 1]
b_grad += -(2 / N) * (y - ((w_current * x) + b_current))
w_grad += -(2 / N) * x * (y - ((w_current * x) + b_current))
new_b = b_current - (learnRate * b_grad)
new_w = w_current - (learnRate * w_grad)
return [new_b, new_w]
# iterate to optimize
def grad_descent_exe(points, starting_b, starting_w, learnRate, num_iterations):
b = starting_b
w = starting_w
for i in range (num_iterations):
b, w = step_grad (b, w, np.array (points), learnRate)
print (b, w)
return [b, w]
def exe():
points = np.genfromtxt ("data.csv", delimiter=",")
# print(points)
learnRate = 0.0001
initial_b = 0 # guessing y-intercept
initial_w = 0 # guessing slope
num_iterations = 1000
print ("Starting grad descent at b = {0}, m = {1}, err = {2}"
.format (initial_b, initial_w, cal_err_linear_given_points (initial_b, initial_w, points)))
[b, w] = grad_descent_exe (points, initial_b, initial_w, learnRate, num_iterations)
print ("After {0} interations b = {1}, m = {2}, err = {3}"
.format (num_iterations, b, w, cal_err_linear_given_points (b, w, points)))
if __name__ == '__main__':
exe ()
