Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

Are all keras layers supported for weight pruning?

See original GitHub issue

Hi, thanks for a very convenient package! Are the layers like LSTM, ConvLSTM2D and TimeDistributed supported for weight pruning? In the other issue, someone had a problem with TimeDistributed layer. If they are not supported out of the box, would inheriting from PrunableLayer and implementing get_prunable_weights() (perhaps as return empty list) fix the issue?

Thanks!

Issue Analytics

State:
Created 4 years ago
Comments:5 (2 by maintainers)

Top GitHub Comments

2reactions

alanchiaocommented, Jul 31, 2019

HI Anuar,

We should be supporting all builtin-Keras layers out-of-the-box, though in the implementation, we need to add them one-by-one (see this). Keras wrappers are a bit different, and in as you can see in the other issue, they aren’t supported yet (BiDirectional and TimeDistributed).

As you suggested, anything missing (either a custom Keras layer … or something that hasn’t been added above yet) can inherit from PrunableLayer (as suggested here).

What Sanjay said is correct also.

1reaction

s36srinicommented, Jul 25, 2019

Based on the _update_mask function in pruning_impl.py, any weight tensor should be prunable as it simply sorts the weights and updates the mask based on the smallest value. Don’t see why this logic should be restricted to a given layer.