GradientDescentMethods.py

def hinge_loss(w, data, labels):
    size = labels.shape
    return np.mean(np.maximum(np.zeros(size), np.ones(size) - labels * np.dot(data, w)))

def hinge_loss_deriv(w, data, labels):
    deriv_sum = np.zeros()
    for x_i, y_i in zip(data, labels):
        if (y_i * np.dot(x_i, w)):
            deriv_sum -= y_i * x_i

    return deriv_sum / labels.shape

def GD(data, label, iters, eta):
    out = np.zeros((labels.shape, t))

    for i in range(iters - 1):
        out[i + 1] = out[i] - eta * hinge_loss_deriv(w, data, labels)
    
    return out