def test_actnorm_invertibility(self):
name = "actnorm"
x, x_mask, _ = self.get_data()
x_inv, logabsdet = glow.actnorm(name,
x,
x_mask,
inverse=False,
init=False)
x_inv_inv, logabsdet_inv = glow.actnorm(name,
x_inv,
x_mask,
inverse=True,
init=False)
self.evaluate(tf.global_variables_initializer())
x, x_inv, x_inv_inv, x_mask, logabsdet, logabsdet_inv = (self.evaluate(
[x, x_inv, x_inv_inv, x_mask, logabsdet, logabsdet_inv]))
diff = x - x_inv_inv
logabsdet_sum = logabsdet + logabsdet_inv
self.assertEqual(x.shape, (BATCH_SIZE, TARGET_LENGTH, N_CHANNELS))
self.assertEqual(x_inv.shape, (BATCH_SIZE, TARGET_LENGTH, N_CHANNELS))
self.assertEqual(x_inv_inv.shape,
(BATCH_SIZE, TARGET_LENGTH, N_CHANNELS))
self.assertTrue(np.allclose(diff, 0.0, atol=1e-5))
self.assertTrue(np.allclose(logabsdet_sum, 0.0, atol=1e-5))
def test_actnorm(self):
_, x_mask, _ = self.get_data()
x = tf.random_normal((BATCH_SIZE, TARGET_LENGTH, N_CHANNELS),
mean=50.0, stddev=10.0, dtype=DTYPE)
x_act, logabsdet = glow.actnorm(
"actnorm", x, x_mask, inverse=False, init=True)
x_act_nopad = tf.boolean_mask(x_act, x_mask)
x_mean, x_var = tf.nn.moments(x_act_nopad, axes=[0])
self.evaluate(tf.global_variables_initializer())
x, x_act, logabsdet, x_mean, x_var = (
self.evaluate([x, x_act, logabsdet, x_mean, x_var]))
self.assertEqual(x_act.shape, (BATCH_SIZE, TARGET_LENGTH, N_CHANNELS))
self.assertEqual(logabsdet.shape, (BATCH_SIZE,))
self.assertTrue(np.allclose(x_mean, 0.0, atol=1e-5))
self.assertTrue(np.allclose(x_var, 1.0, atol=1e-5))