Getting error when trying to wrap a model with a tf keras Normalization layer

[fabiodimarco]

Getting error when trying to wrap a model with a tf keras Normalization layer

[fabiodimarco]

Hi Tanuj,

could you provide me a code to replicate the issue? Instead of using the real database you can also use a randomly generated one with same input/output dimensions.

[tk3016]

Hi

Thanks for your prompt response.
I am attaching a different code with the same issue I am facing which is based on the regression tutorial on Tensorflow's website.
tf_regression_tutorial_with_error.zip

[fabiodimarco]

Hi Tanuj,

I ran your notebook, the error is not due to the ModelWrapper, but it is caused by this line of code:
tf.keras.models.clone_model(dnn_model_shallow)

Actually you do not need to make a clone of the model, you can just pass it directly to the ModelWrapper.

model_wrapper = lm.ModelWrapper(dnn_model_shallow)

model_wrapper.compile(
    optimizer=tf.keras.optimizers.SGD(learning_rate=0.1),
    loss=lm.MeanSquaredError())

history = model_wrapper.fit(
    train_features,
    train_labels,
    validation_split=0.2,
    verbose=1,
    epochs=100
)

Let me know if this solves your problem.

Best,
Fabio

[tk3016]

Hi Fabio,

Many thanks for the suggestion. It works!
I was having a look at the curve fitting example and similarly trying to clone the model which is what confused me.

Thanks a lot again for all your help.

Best wishes,
Tanuj