Cannot convert Keras saved model

System information

• Ubuntu 18.04 WSL
• tensorflow 2.5.0

Describe the problem

Model convertion using tensorflowjs_converter --input_format=keras_saved_model ./model_autokeras ./converted_model throws following error:

RuntimeError: Unable to restore a layer of class Custom>MultiCategoryEncoding. Layers of class Custom>MultiCategoryEncoding require that the class be provided to the model loading code, either by registering the class using @keras.utils.register_keras_serializable on the class def and including that file in your program, or by passing the class in a keras.utils.CustomObjectScope that wraps this load call.

My model has been generated using Autokeras StructuredDataRegressor.

Provide the exact sequence of commands / steps that you executed before running into the problem

This can be reproduced by following this tutorial: https://autokeras.com/tutorial/structured_data_regression/

And then run tensorflowjs_converter --input_format=keras_saved_model ./model_autokeras ./converted_model

[wingman-jr-addon]

Looks like you might have a custom layer that is not supported “natively” in Tensorflow.js. Have you tried changing the output format to --output_format=tfjs_graph_model?

[google-ml-butler[bot]]

Are you satisfied with the resolution of your issue? Yes No