def train(X, y, W, b, tolerance = 1e-10):
Jhist = []
i = 1
while True:
    m = y.size

    z = X @ W + b
    a = sigmoid(z)

    HessW = (1/m) * (a.T @ (1-a)) * (X.T @ X)
    gradW = (1/m) * (X.T @ (a-y))

    secDer_b = (1/m) * (a.T @ (1-a))
    frstDer_b = (1/m) * np.sum(a-y)

    W -= 0.00001*(np.linalg.inv(HessW) @ gradW)
    b -= 0.00001*(secDer_b/frstDer_b)

    J = cost(X, y, W, b)

    if i != 1 and abs(Jhist[-1] - J) < tolerance:
        print('converged!!!')
        break

    Jhist.append(J)

    if i%100 == 0:
        print("Iteration:: {} - Cost:: {}".format(i, J))

    i += 1

return W, b, Jhist