Something wrong with the creation of a Keras model with probabilistic layers.

[gutilacerda]

I am applying the use of Keras models that utilize probabilistic layers. The model below, which used to work normally, now presents the following error:

inputs = tf.keras.Input(shape=(X_train.shape[1],)) x = layers.Dense(500, activation='relu')(inputs) x2 = layers.Dense(350, activation='relu')(x) x3 = layers.Dense(250, activation='relu')(x2) x4 = layers.Dense(150, activation='relu')(x3) x5 = layers.Dense(50, activation='relu')(x4) x6 = layers.Dense(15, activation='relu')(x5) y_variational = tfpl.DenseVariational(4, make_prior_fn=prior, make_posterior_fn=posterior, kl_weight=1/X_train.shape[0], activation='tanh')(x6)

distribution_params = layers.Dense(units=216*2)(y_variational) outputs = tfpl.IndependentNormal(216)(distribution_params)

PBNN3 = tf.keras.Model(inputs=inputs, outputs=outputs)

---

AttributeError Traceback (most recent call last) in <cell line: 8>() 6 x5 = layers.Dense(50, activation='relu')(x4) #15 7 x6 = layers.Dense(15, activation='relu')(x5) ----> 8 y_variational = tfpl.DenseVariational(4, make_prior_fn=prior, make_posterior_fn=posterior, kl_weight=1/X_train.shape[0], activation='tanh')(x6) 9 #y_variational = layers.Dense(10, activation='relu')(x) 10 distribution_params = layers.Dense(units=216*2)(y_variational)

1 frames /usr/local/lib/python3.10/dist-packages/tf_keras/src/engine/input_spec.py in assert_input_compatibility(input_spec, inputs, layer_name) 249 ) 250 if spec.min_ndim is not None: --> 251 ndim = x.shape.rank 252 if ndim is not None and ndim < spec.min_ndim: 253 raise ValueError(

AttributeError: 'tuple' object has no attribute 'rank'

Something is not working with TensorFlow version 2.17.0 and TensorFlow Probability version 0.24. Still works for Tensorflow: 2.15.1 and TensorFlow Probability: 0.22.1.

Thank you!

[kittmiller25]

I was having a similar issue and just switched all of my tf.keras calls to tf_keras and it solved it. I believe it is regarding the Keras 2 vs Keras 3 discrepancy in the 0.24 release notes.

https://github.com/tensorflow/probability/releases NOTE: In TensorFlow 2.16+, tf.keras (and tf.initializers, tf.losses, and tf.optimizers) refers to Keras 3. TensorFlow Probability is not compatible with Keras 3 -- instead TFP is continuing to use Keras 2, which is now packaged as tf-keras and tf-keras-nightly and is imported as tf_keras. When using TensorFlow Probability with TensorFlow, you must explicitly install Keras 2 along with TensorFlow (or install tensorflow-probability[tf] or tfp-nightly[tf] to automatically install these dependencies.)

The one that wasn't mentioned but I think is the one that actually got my code working again was the Model itself so: tf.keras.Model -> tf_keras.Model I went wholesale and did it to all of my layers, activation functions, optimizers, etc. but I'm not sure if that was necessary

[jhn-nt]

is there a roadmap to make TFP support Keras 3?

Thanks for all the amazing work! Giovanni