Feature/robust kge

[Akash-bu]

Implemented UNITEI loss function for pattern aware and noise resilient embeddings.

Request to review the code and provide feedback.

[alkidbaci]

Hello @Akash-bu

You are trying to add 8 million new lines into the repository.

The PRs should contain only changes that improve our library and not experiment or dataset files.

Bulky PRs (i.e. PRs with a large amount of code changes) will not be considered.

If you believe you have code enhancements that may be useful for other users of dice-embeddings, please make your PR more concise as to my instructions otherwise we will have to discard this PR.

[Demirrr]

Thank you @alkidbaci .

@Akash-bu
you might want to know that we do not merge the changes into dev unless the changes

1. fixes a bug,
2. improves the efficiency/readability or
3. implements an algorithm accepted in a conference

[Akash-bu]

Hi @alkidbaci and @Demirrr

Thank you for reviewing and for the feedback.

Just to clarify — this PR is not intended for the dev branch.
It’s meant to extend the feature/label-relaxation branch, which is part of our ongoing internal research work.

I’ve cleaned up the branch by removing all dataset and experimental output files so that it now only contains the core code updates (unitei_loss.py, base_model.py, and .gitignore).

[alkidbaci]

Your changes now consist only of 3 lines in .gitignore. Make sure you commit and push your changes for the other 2 scripts that you mention: unitei_loss.py and base_model.py

[Akash-bu]

Both files are now committed.