tf.keras compatibility: convert multi-label-classification to Keras Core

[Akorex]

for tf.keras backwards compatibility

In my attempt to port Large-scale multi-label classification to keras_core I hit an issue with the Inference Section. The example from keras.io created a new Inference model after training to include the TextVectorization preprocessing layer so inference can be done by passing strings directly to the model.

model_for_inference = keras.Sequential([text_vectorizer, shallow_mlp_model])

This didn't work for keras_core imports and the attempt to port the example didn't include the particular step from the example.

PR: keras-team/keras-core#538

Colab Notebook: https://colab.research.google.com/drive/1kR7W4BJhtDlA5Jx_e4EwUM2FnzKP-D58#scrollTo=xmf6rTLL0G9H

[grasskin]

Thank you for opening this issue! I'm getting access denied on the colab notebook, could you change the permissions/send a different link?

[Akorex]

I have updated the permissions to the notebook and you should have access to it now.

[fchollet]

The notebook you shared doesn't seem to include the line model_for_inference = .... Can you post a shorter, more focused reproduction notebook?

[Akorex]

I have added the line model_for_inference = ... in the new notebook. However, I hit an issue with layers.StringLookup() using the keras_code import which didn't exist with the tf.keras import

Notebook : https://colab.research.google.com/drive/134sncHaoAaV9xeV7fOHVDZMVu12rAUVI#scrollTo=iwVJj6dK0G9N