gradient descent 직접 구현하기 (linear regression)

import numpy as np

# linear regression
x = np.random.rand(1000)
y = x * 2 + np.random.randn()
print(x, y)

# parameter
w = 0.0
b = 0.0

# hyperparameter
lr = 0.01
epochs = 100000

def linear_regression(w1:float, b1:float, x1:float) -> float:
    # TODO
    return w1 * x1 + b1
    
def loss_function(x2:np.array, y2:np.array, w2:float, b2:float) -> float:
    # TODO
    yhat = w2 * x2 + b2
    return np.divide(np.sum((y2 - yhat) ** 2), x.shape[0])


def gradient_descent(x3:np.array, y3:np.array, w3:float, b3:float, lr:float) -> tuple:
    # TODO
    dw = 0.0
    db = 0.0
    N = x3.shape[0]
    for xi, yi in zip(x3, y3):
        dw += -2 * xi * (yi - linear_regression(w3, b3, xi))
        db += -2 * (yi - linear_regression(w3, b3, xi))
    w3 -= dw * lr * (1/N)
    b3 -= db * lr * (1/N)
    
    return w3, b3

for i in range(epochs):
    w, b = gradient_descent(x,y,w,b,lr)
    yhat = w * x + b
    loss = np.divide(np.sum((y - yhat) ** 2), x.shape[0])
    print(f"{i} weight {w:.4f} bias {b:.4f} loss {loss:.4f}")