Error while importing a Keras Model with a TFHub layer

[guidiandrea]

Hello,

I tried importing a Keras model saved in h5 format with the Model.load_keras API, but I got the following error:
Unknown layer: KerasLayer

Which obviously refers to the tensorflow-hub layer. Does this mean that tfhub models can't be used with bigDL?

[hkvision]

Can you provide with more error message and some code segments?

[guidiandrea]

A minimal working example:

import tensorflow as tf
import tensorflow_hub as hub
hub_layer = hub.KerasLayer("https://tfhub.dev/google/tf2-preview/gnews-swivel-20dim/1", output_shape=[20],
                           input_shape=[], dtype=tf.string)
model = tf.compat.v1.keras.Sequential()
model.add(hub_layer)
model.summary()

model.save("text.h5",save_format='h5')
#I can't reload the model
bigdl_model = Model.load_keras(hdf5_path="text.h5")

And here you can find the error message I got

[hkvision]

Model.load_keras is only tested on Keras 1.2.2.

[guidiandrea]

Thank you @hkvision

Do you have other workarounds to load a tfhub model (other APIs which I don't know...)?

[hkvision]

You can take a look at TFPark in Analytics Zoo. For the support of TF2, we are still working in progress. TFPark supports TF 1.15 and the corresponding tf.keras.