def test_encode_image_rVAE(latent_dim):
large_image = np.random.randn(2, 64, 64)
window_size = (16, 16)
patches_data = np.random.randn(32, *window_size)
v = rVAE(window_size, latent_dim)
v.fit(patches_data, training_cycles=2)
img, img_encoded = v.encode_images(large_image)
cropped_dim = 64 - window_size[0] + 1
assert_equal(img.shape, (2, cropped_dim, cropped_dim))
assert_equal(img_encoded.shape,
(2, cropped_dim, cropped_dim, latent_dim + 3))
def test_encoding_rVAE(translation, latent_dim, encoded_dim):
input_dim = (28, 28)
data = gen_image_data()
v = rVAE(input_dim,
latent_dim,
translation=translation,
numhidden_encoder=16,
numhidden_decoder=16)
v.fit(data, training_cycles=2)
z = v.encode(data)
assert_equal(len(z), 2)
assert_equal(z[0].shape[-1], encoded_dim)
assert_equal(z[1].shape[-1], encoded_dim)
def test_decoding_rVAE(conv_encoder, conv_decoder, translation, latent_dim):
input_dim = (28, 28)
data = gen_image_data()
v = rVAE(input_dim,
latent_dim,
translation=translation,
numhidden_encoder=16,
numhidden_decoder=16)
v.fit(data, training_cycles=2)
z_sample = np.zeros((latent_dim))[None]
decoded = v.decode(z_sample)
assert_equal(decoded.shape[1:], input_dim)
assert_(np.sum(decoded) != 0)