Fix LMU swapping behaviour during training #28
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
In some situations (e.g. `.fit`) the layer gets called with a mix of defined and undefined input shapes, which messes up the autoswapping logic. This changes it so that the swap type is fixed at build time.
drasmuss
force-pushed
the
fix-swap-fit
branch
from
cf24f8d to
ebc9b08
Compare
gsmalik
approved these changes
Nov 16, 2020
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
In some situations (e.g. calling
.fitwithNonefor the time dimension) the layer gets called with a mix of defined and undefined input steps, which messes up the autoswapping logic inkeras_lmu.LMU(between the RNN and FFT implementations). This changes it so that the swap type is fixed at build time.Fixes #27