hub.KerasLayer: add a progress bar when downloading/building layer

[galeone]

I'm creating a simple model to test the transfer learning capabilities offered by hub.KerasLayer in this way:

m = tf.keras.Sequential([
    hub.KerasLayer("https://tfhub.dev/google/tf2-preview/inception_v3/feature_vector/2", output_shape=[2048],
                   trainable=False),
    tf.keras.layers.Dense(5)
])

The code runs correctly, but every time I build the model it freezes without giving any output until it downloads the module (I guess).

I guess it could be useful to add a progress bar (optional?) when the module is downloading and preparing the layer.

Info:

• tensorflow-hub: 0.4
• tensorflow: 2.0 nightly

[captain-pool]

Hey @vbardiovskyg Can I work on this?

[vbardiovskyg]

Hi Paolo and Adrish,

there is already a progress bar implemented and can be turned on by setting an environment variable defined here: https://github.com/tensorflow/hub/blob/d3de17ac3f7f82336a3812f58729676a46be22d7/tensorflow_hub/resolver.py#L47.

import os
os.environ['TFHUB_DOWNLOAD_PROGRESS'] = "1"

[galeone]

oh cool, thanks! It is something official enabling it in this way or it is a hack?

(asking because I'm writing a book and I'm writing about hub.KerasLayer and it would be nice to write something official about how to enable the progress bar)

[captain-pool]

oh cool, thanks! It is something official enabling it in this way or it is a hack?

(asking because I'm writing a book and I'm writing about hub.KerasLayer and it would be nice to write something official about how to enable the progress bar)

I went through the resolver.py looks like it's official.
https://github.com/tensorflow/hub/blob/54a242de6e148d0836c4a58308b07bbe79c274b6/tensorflow_hub/resolver.py#L132-L144

[vbardiovskyg]

It's a public API, although not part of the functional one and maybe not very well documented.

[galeone]

Perfect, thank you for the info!

[rmothukuru]

Closing this issue as it has been resolved.