Fix CNN tensor multihot input - #1208
Open
fbonc wants to merge 2 commits into
Open
Conversation
CNN.forward hardcoded a 3-D expectation for spatial_dim=1 features, but MultiHotProcessor and 1D TensorProcessor inputs embed to [batch, embedding_dim] with no sequence axis. Treat these as a length-1 sequence.
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Issue
CNN’s one-dimensional convolution path expects embedded features shaped [batch, sequence_length, embedding_dim]. Sequence processors produce that shape, but MultiHotProcessor and one-dimensional TensorProcessor inputs produce one embedding per sample, shaped [batch, embedding_dim]. Because the sequence axis is absent, CNN rejects these otherwise-supported inputs as two-dimensional tensors.
Fix
For the Conv1d path, normalize a [batch, embedding_dim] embedding to [batch, 1, embedding_dim] before validating and permuting dimensions. The existing permutation then gives Conv1d its required [batch, embedding_dim, 1] input. Inputs that already contain a sequence dimension are unchanged.
Notes
The regression test runs CNN with both a multi-hot feature and a one-dimensional tensor feature and verifies that forward and backward propagation complete successfully.