NumPy-linear_reg-3.py

def gradient_descent(X, y, w, b, learning_rate):
    dw = -2 * np.sum(X * (y - w * X - b)) # ∂e/∂w
    db = -2 * np.sum(y - w * X - b)       # ∂e/∂b
    w_new = w - learning_rate * dw        # minus sign since we are minizing e
    b_new = b - learning_rate * db
    return w_new, b_new

def get_loss(X,y,w,b):
    return (y - w * X - b).T @ (y - w * X - b)   # square loss, 
    # .T and @ denote transpose and matrix multiplication resp.
    

learning_rate = 0.000001
max_epoch = 500
w, b = -1,0

for epoch in range(1,max_epoch+1):
    w,b = gradient_descent(X, y, w, b, learning_rate)
    
    if epoch % 50 == 0:
        print(f'{get_loss(X,y,w,b):.0f}')

if b > 0:
    print(f'y = {w:.2f} x + {b:.2f}')
else:
    print(f'y = {w:.2f} x - {-b:.2f}')