Last active
June 19, 2019 18:47
-
-
Save keithmgould/27b9ab35592adf5282c96fea25c17031 to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
# toy example from https://www.kdnuggets.com/2017/04/simple-understand-gradient-descent-algorithm.html | |
import numpy as np | |
import random | |
import matplotlib.pyplot as plt | |
LEARNING_RATE_M = 1e-8 | |
LEARNING_RATE_B = 1e-2 | |
# sq ft , price | |
HISTORIC_DATA = [ | |
[1400, 245000], | |
[1600, 312000], | |
[1700, 279000], | |
[1875, 308000], | |
[1100, 199000], | |
[1550, 219000], | |
[2350, 405000], | |
[2450, 324000], | |
[1425, 319000], | |
[1700, 255000]] | |
N = len(HISTORIC_DATA) | |
def determine_error(m, b): | |
error_sum = 0 | |
for x, y in HISTORIC_DATA: | |
y_hat = m * x + b | |
error_sum += .5 * ( y_hat - y) ** 2 | |
return error_sum / float(N) | |
def calculate_gradients(m, b): | |
grad_m = 0 | |
grad_b = 0 | |
for x, y in HISTORIC_DATA: | |
y_hat = m * x + b | |
grad_b += y_hat - y | |
grad_m += (y_hat - y) * x | |
grad_m = grad_m / N | |
grad_b = grad_b / N | |
return grad_m, grad_b | |
# f(x) = mx + b | |
def main(): | |
m = random.randint(1,100) # slope | |
b = random.randint(20,25) # bias | |
plt.subplots(5,1, figsize=(6,10)) | |
plot = 1 | |
for i in range(40000): | |
total_error = determine_error(m,b) | |
# print("total error: {}. a:{}, b:{}".format(total_error, m, b)) | |
grad_m, grad_b = calculate_gradients(m,b) | |
# print("{},{}".format(grad_m, grad_b)) | |
b = b - LEARNING_RATE_B * grad_b | |
m = m - LEARNING_RATE_M * grad_m | |
hd = np.array(HISTORIC_DATA) | |
x = hd[:,0] | |
y = hd[:,1] | |
if i % 4000 == 0: | |
plt.subplot(10, 1, plot) | |
plt.scatter(x,y) | |
# plots ground truth line of best fit: | |
plt.plot(np.unique(x), np.poly1d(np.polyfit(x, y, 1))(np.unique(x))) | |
# plots estimated line of best fit: | |
plt.plot([1000,3000],[m*1000+b,m*3000+b]) | |
plot += 1 | |
print("m: {}, b: {}".format(m,b)) | |
plt.show() | |
if __name__ == '__main__': | |
main() |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment