def test_gradient_check(self):
x = dz.Variable(np.random.rand(1))
y = dz.square(x)
y.backward()
num_grad = numerical_diff(dz.square, x)
flg = np.allclose(x.grad, num_grad)
self.assertTrue(flg)
# -*- coding: utf-8 -*-
import dezero
import numpy as np
if __name__ == "__main__":
A = dezero.Square()
B = dezero.Exp()
C = dezero.Square()
# forward
x = dezero.Variable(np.array(0.5))
a = A(x)
b = B(a)
y = C(b)
# Backwards
y.grad = np.array(1.0)
y.backward()
print(x.grad)
def test_backward(self):
x = dz.Variable(np.array(2.0))
y = dz.exp(x)
y.backward()
expected = np.exp(2.0)
self.assertEqual(x.grad, expected)
def numerical_diff(f, x, eps=1e-4):
x0 = dz.Variable(x.data - eps)
x1 = dz.Variable(x.data + eps)
y0 = f(x0)
y1 = f(x1)
return (y1.data - y0.data) / (2 * eps)
def test_forward(self):
x = dz.Variable(np.array(2.0))
y = dz.exp(x)
expected = np.exp(2.0)
self.assertEqual(y.data, expected)