nasa-petal / PeTaL-labeller

The PeTaL labeler labels journal articles with biomimicry functions.
https://petal-labeller.readthedocs.io/en/latest/
The Unlicense
6 stars 4 forks source link

First prototype of auto labeler #3

Closed shrutix closed 3 years ago

shrutix commented 3 years ago

Summary

Prototype of auto labeler for articles and publications. Uses transformers and BERT architecture (fine-tuned) with PyTorch to classify wikipedia articles. Current classification accuracy: 11%

Related Issues

Backwards incompatibilities

Does this PR cause any incompatibilites with previous versions? Otherwise, None.

New Dependencies

Does this PR require the use of a new dependency? Otherwise, None.

pjuangph commented 3 years ago
  1. Shruthi in the demo, i remember you had some technical problems with the code. Have you addressed that?

  2. You reference both tensorflow and pytorch? do you actually use tensorflow? Can you remove unused references? Also remove lengthy print statements.

shrutix commented 3 years ago

@pjuangph turns out those were some issues with the colab gpu, it's running smoothly now I removed unecessary print statements and I think I use tensorflow in the beginning to check that the gpu is available.