Improved Transformer example #23
Open
Conversation
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
…n. Add sentiment analysis task and permutation parity task.
…that are unseperable without positional encoding
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
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Outdated suggestions cannot be applied.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.
The current synthetic test is positionally invariant, and thus fails to test the positional encoding. I've added a Symmetry Detection Test that converges to 98% accuracy with sinusoidal PE and 50% without. I also added an optional real-world sentiment analysis test, though it requires Hugging Face's datasets and transformers libraries. It's completely optional.
The model expected one-hot encoded vectors which it would pass through a dense layer. This is non-standard and quite slow for real-world use-cases. I fixed it by replacing the dense layer with a standard embedding layer. The model also used a non-standard mean pooling which can destroy PE-related gradients. I implemented a BERT-style CLS token.