autoencoder.fit doesnt work becaue of a ValueError

Question

I don't understand what my problem is. It should work, if only because its the standard autoenoder from the tensorflow documentation. this is the error

line 64, in call
    decoded = self.decoder(encoded)
ValueError: Exception encountered when calling Autoencoder.call().

Invalid dtype: <property object at 0x7fb471cc1c60>

Arguments received by Autoencoder.call():
  • x=tf.Tensor(shape=(32, 28, 28), dtype=float32)

and this is my code

(x_train, _), (x_test, _) = fashion_mnist.load_data()

x_train = x_train.astype('float32') / 255.
x_test = x_test.astype('float32') / 255.

print (x_train.shape)
print (x_test.shape)

class Autoencoder(Model):
  def __init__(self, latent_dim, shape):
    super(Autoencoder, self).__init__()
    self.latent_dim = latent_dim
    self.shape = shape
    self.encoder = tf.keras.Sequential([
      layers.Flatten(),
      layers.Dense(latent_dim, activation='relu'),
    ])
    self.decoder = tf.keras.Sequential([
      layers.Dense(tf.math.reduce_prod(shape), activation='sigmoid'),
      layers.Reshape(shape)
    ])

  def call(self, x):
    encoded = self.encoder(x)
    print(encoded)
    decoded = self.decoder(encoded)
    print(decoded)
    return decoded


shape = x_test.shape[1:]
latent_dim = 64
autoencoder = Autoencoder(latent_dim, shape)

autoencoder.compile(optimizer='adam', loss=losses.MeanSquaredError())

autoencoder.fit(x_train, x_train,
                epochs=10,
                shuffle=True,
                validation_data=(x_test, x_test))

I tried to change the database and also tried different shapes

Not really. You may have a corrupt installation, or incompatible libraries (e.g. Keras vs Tensorflow). Is this the whole error traceback? It's unfortunately not very useful. — xdurch0
– xdurch0, Commented Mar 17, 2024 at 18:38

0not · Accepted Answer · 2024-03-17 22:10:42Z

I ran into the same error when trying to get this example working with Keras 3. The invalid dtype error is because the Dense layer in the decoder expects a positive integer, but reduce_prod returns a scalar tensor. You must extract the scalar value with, e.g. numpy():

layers.Dense(tf.math.reduce_prod(shape).numpy(), activation='sigmoid')

After fixing that error, I ran into a problem with batch sizes (the model in the example doesn't expect a batch dimension), which I fixed with an initial Input layer in the encoder. Here is my Autoencoder model converted to Keras 3:

class Autoencoder(keras.Model):

  def __init__(self, latent_dim, shape):
    super().__init__()

    self.latent_dim = latent_dim
    self.shape = shape

    self.encoder = keras.Sequential([
      keras.Input(shape),
      keras.layers.Flatten(),
      keras.layers.Dense(latent_dim, activation='relu'),
    ])

    self.decoder = keras.Sequential([
      keras.layers.Dense(keras.ops.prod(shape).numpy(), activation='sigmoid'),
      keras.layers.Reshape(shape)
    ])

  def call(self, x):
    encoded = self.encoder(x)
    decoded = self.decoder(encoded)

    return decoded

shape = x_test.shape[1:]
latent_dim = 64
autoencoder = Autoencoder(latent_dim, shape)

Collectives™ on Stack Overflow

autoencoder.fit doesnt work becaue of a ValueError

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related