def test_vae_predictions(self):
# Encoder.
encoder = models.Sequential()
encoder.add(layers.Dense(20, activation="relu", input_shape=(100, )))
# Decoder.
decoder = models.Sequential()
decoder.add(layers.Dense(100, activation="sigmoid", input_shape=(2, )))
# Variational Autoencoder.
vae = ngdlmodels.VAE(encoder=encoder, decoder=decoder, latent_dim=2)
prediction = vae.predict(np.random.random((1, 100)))
assert prediction.shape == (
1, 100), "Unexpected shape " + str(prediction.shape)
prediction = vae.predict_reconstruct_from_samples(
np.random.random((1, 100)))
assert prediction.shape == (
1, 100), "Unexpected shape " + str(prediction.shape)
prediction = vae.predict_embed_samples_into_latent(
np.random.random((1, 100)))
assert prediction.shape == (
1, 2), "Unexpected shape " + str(prediction.shape)
prediction = vae.predict_reconstruct_from_latent(
np.random.random((1, 2)))
assert prediction.shape == (
1, 100), "Unexpected shape " + str(prediction.shape)
def _build_dense(type, input_shape, latent_dim, hidden_units=[], hidden_activation="relu"):
# Some useful variables.
input_size = np.prod(input_shape)
hidden_units_reverse = hidden_units[::-1]
# Create the encoder.
encoder_input = layers.Input(shape=input_shape)
encoder_output = layers.Reshape((input_size,))(encoder_input)
for hidden in hidden_units:
encoder_output = layers.Dense(hidden, activation=hidden_activation)(encoder_output)
encoder_output = layers.Dense(latent_dim, activation=hidden_activation)(encoder_output)
encoder = models.Model(encoder_input, encoder_output)
# Create the decoder.
decoder_input = layers.Input(shape=(latent_dim,))
decoder_output = decoder_input
for hidden in hidden_units_reverse:
decoder_output = layers.Dense(hidden, activation=hidden_activation)(decoder_output)
decoder_output = layers.Dense(input_size, activation="sigmoid")(decoder_output)
decoder_output = layers.Reshape(input_shape)(decoder_output)
decoder = models.Model(decoder_input, decoder_output)
# Create the autoencoder.
if type == "ae":
ae = ngdlmodels.AE(encoder, decoder)
return ae
elif type == "vae":
vae = ngdlmodels.VAE(encoder, decoder, latent_dim=latent_dim)
return vae
def test_issue_1(self):
input_shape = (28, 28)
latent_dim = 128
encoder_input = layers.Input(shape=input_shape)
encoder_output = encoder_input
encoder_output = layers.Reshape(input_shape + (1, ))(encoder_input)
encoder_output = layers.Conv2D(16, (3, 3),
activation="relu",
padding="same")(encoder_output)
encoder_output = layers.MaxPooling2D((2, 2),
padding="same")(encoder_output)
encoder_output = layers.Conv2D(8, (3, 3),
activation="relu",
padding="same")(encoder_output)
encoder_output = layers.MaxPooling2D((2, 2),
padding="same")(encoder_output)
encoder_output = layers.Conv2D(8, (3, 3),
activation="relu",
padding="same")(encoder_output)
encoder_output = layers.MaxPooling2D((2, 2),
padding="same")(encoder_output)
encoder_output = layers.Flatten()(encoder_output)
encoder = models.Model(encoder_input, encoder_output)
# Create the decoder.
decoder_input = layers.Input(shape=(latent_dim, ))
decoder_output = decoder_input
#decoder_output = layers.Dense(128, activation="relu")(decoder_output)
decoder_output = layers.Reshape((4, 4, 8))(decoder_output)
decoder_output = layers.Conv2D(8, (3, 3),
activation="relu",
padding="same")(decoder_output)
decoder_output = layers.UpSampling2D((2, 2))(decoder_output)
decoder_output = layers.Conv2D(8, (3, 3),
activation="relu",
padding="same")(decoder_output)
decoder_output = layers.UpSampling2D((2, 2))(decoder_output)
decoder_output = layers.Conv2D(16, (3, 3),
activation="relu")(decoder_output)
decoder_output = layers.UpSampling2D((2, 2))(decoder_output)
decoder_output = layers.Conv2D(1, (3, 3),
activation="sigmoid",
padding="same")(decoder_output)
decoder_output = layers.Reshape((28, 28))(decoder_output)
decoder = models.Model(decoder_input, decoder_output)
# Create the VAE.
vae = ngdlmodels.VAE(encoder, decoder, latent_dim=latent_dim)
vae.compile(optimizer='adadelta', loss="binary_crossentropy")
vae.summary()
# Train.
print("Train...")
history = vae.fit(x_input_train,
x_input_train,
epochs=100,
batch_size=32,
shuffle=True,
validation_data=(x_input_test, x_input_test))
def build_model(self):
"""
Build the VAE gluing the
encoder and decoder using
the ngdlmodels.
"""
encoder = self.encoder
decoder = self.decoder
latent_dim = self.latent_dim
self.model = ngdlmodels.VAE(encoder, decoder, latent_dim=latent_dim)
def test_vae_sequential(self):
# Encoder.
encoder = models.Sequential()
encoder.add(layers.Dense(20, activation="relu", input_shape=(100, )))
# Decoder.
decoder = models.Sequential()
decoder.add(layers.Dense(100, activation="sigmoid", input_shape=(2, )))
# Variational Autoencoder.
vae = ngdlmodels.VAE(encoder=encoder, decoder=decoder, latent_dim=2)
self.assert_load_save_fine(vae)
def test_vae_functional(self):
# Encoder.
encoder_input = layers.Input(shape=(100,))
encoder_output = encoder_input
encoder_output = layers.Dense(20, activation="relu")(encoder_output)
encoder = models.Model(encoder_input, encoder_output)
# Decoder.
decoder_input = layers.Input(shape=(2,))
decoder_output = layers.Dense(20, activation="relu")(decoder_input)
decoder_output = layers.Dense(100, activation="sigmoid")(decoder_output)
decoder = models.Model(decoder_input, decoder_output)
# Variational Autoencoder.
vae = ngdlmodels.VAE(encoder=encoder, decoder=decoder, latent_dim=2)
vae.compile(optimizer="rmsprop", loss="mse")
self.assert_load_save_fine(vae)
# Create the encoder.
encoder_input = layers.Input(shape=(28, 28))
encoder_output = layers.Reshape((28 * 28, ))(encoder_input)
encoder_output = layers.Dense(512, activation="relu")(encoder_output)
encoder = models.Model(encoder_input, encoder_output)
# Create the decoder.
decoder_input = layers.Input(shape=(latent_dim, ))
decoder_output = layers.Dense(512, activation="relu")(decoder_input)
decoder_output = layers.Dense(28 * 28, activation="sigmoid")(decoder_output)
decoder_output = layers.Reshape((28, 28))(decoder_output)
decoder = models.Model(decoder_input, decoder_output)
# Create the autoencoder.
vae = ngdlmodels.VAE(encoder, decoder, latent_dim=latent_dim)
vae.compile(optimizer='adadelta', reconstruction_loss="binary_crossentropy")
vae.summary()
# Train.
print("Train...")
history = vae.fit(x_input_train,
x_input_train,
epochs=100,
batch_size=32,
shuffle=True,
validation_data=(x_input_test, x_input_test))
# Evaluate.
print("Evaluate...")
loss = vae.evaluate(x_input_test, x_input_test)
def define_VAE_architecture(latent_dim=10):
''' Defines a variational autoenconder with a
convolutional neural network architecture
for our 128 x 128 images.
Returns the model.
'''
### Definition of NN architecture
latent_dim = latent_dim
# Encoder.
encoder_input = layers.Input(shape=(128, 128, 1))
encoder_output = layers.Conv2D(
32, (3, 3),
kernel_initializer=initializers.lecun_normal(seed=0),
activation="relu",
padding="same")(encoder_input)
encoder_output = layers.MaxPooling2D((2, 2),
padding="same")(encoder_output)
encoder_output = layers.Conv2D(
32, (3, 3),
kernel_initializer=initializers.lecun_normal(seed=0),
activation="relu",
padding="same")(encoder_output)
encoder_output = layers.MaxPooling2D((2, 2),
padding="same")(encoder_output)
encoder_output = layers.Flatten()(encoder_output)
encoder_output = layers.Dense(
1024,
activation="relu",
kernel_initializer=initializers.lecun_normal(seed=0))(encoder_output)
encoder_output = layers.Dense(
64,
activation="relu",
kernel_initializer=initializers.lecun_normal(seed=0))(encoder_output)
encoder = models.Model(encoder_input, encoder_output)
# Decoder.
decoder_input = layers.Input(shape=(latent_dim, ))
decoder_output = layers.Dense(
64,
activation="relu",
kernel_initializer=initializers.lecun_normal(seed=0))(decoder_input)
decoder_output = layers.Dense(
1024,
activation="sigmoid",
kernel_initializer=initializers.lecun_normal(seed=0))(decoder_output)
decoder_output = layers.Dense(
32 * 32 * 32,
activation="sigmoid",
kernel_initializer=initializers.lecun_normal(seed=0))(decoder_output)
decoder_output = layers.Reshape((32, 32, 32))(decoder_output)
decoder_output = layers.Conv2D(
32, (3, 3),
kernel_initializer=initializers.lecun_normal(seed=0),
activation="relu",
padding="same")(decoder_output)
decoder_output = layers.UpSampling2D((2, 2))(decoder_output)
decoder_output = layers.Conv2D(
32, (3, 3),
kernel_initializer=initializers.lecun_normal(seed=0),
activation="relu",
padding="same")(decoder_output)
decoder_output = layers.UpSampling2D((2, 2))(decoder_output)
decoder_output = layers.Conv2D(
1, (3, 3),
kernel_initializer=initializers.lecun_normal(seed=0),
activation="sigmoid",
padding="same")(decoder_output)
decoder = models.Model(decoder_input, decoder_output)
#autoencoder
vae = ngdlmodels.VAE(encoder, decoder, latent_dim=latent_dim)
return vae
