[Python] Model definition in Keras tutorial "Probabilistic Bayesian Neural Networks" leads to...

I am trying to run the keras-tutorial Probabilistic Bayesian Neural Networks to get an understanding of Bayesian neural networks (BNN). The tutorial contains a google-colab notebook, so you can run it directly within the browser. However, when I try to train the BNN for the first time (chapter Train BNN with a small training subset), it throws this error:

AttributeError Traceback (most recent call last)
<ipython-input-11-c94aabf54886> in <cell line: 5>()
3 small_train_dataset = train_dataset.unbatch().take(train_sample_size).batch(batch_size)
4
----> 5 bnn_model_small = create_bnn_model(train_sample_size)
6 run_experiment(bnn_model_small, mse_loss, small_train_dataset, test_dataset)

2 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'


The error message probably refers to the model definition in the cell before:

Model Definition


def create_bnn_model(train_size):
inputs = create_model_inputs()
features = keras.layers.concatenate(list(inputs.values()))
features = layers.BatchNormalization()(features)

# Create hidden layers with weight uncertainty using the DenseVariational layer.
for units in hidden_units:
features = tfp.layers.DenseVariational(
units=units,
make_prior_fn=prior,
make_posterior_fn=posterior,
kl_weight=1 / train_size,
activation="sigmoid",
)(features)

# The output is deterministic: a single point estimate.
outputs = layers.Dense(units=1)(features)
model = keras.Model(inputs=inputs, outputs=outputs)
return model


The code to start the training is this:
Start training

num_epochs = 500
train_sample_size = int(train_size * 0.3)
small_train_dataset = train_dataset.unbatch().take(train_sample_size).batch(batch_size)

bnn_model_small = create_bnn_model(train_sample_size)
run_experiment(bnn_model_small, mse_loss, small_train_dataset, test_dataset)


I tried wrapping the features - tensor into a tf.keras.input - layer, but it did not solve the problem

Continue reading...