Customize dataset and classes/label for training

[JanRodriguez]

It would be great if you could customize a dataset to use for training, for instance with CoNLL files (similar to this).

[richardjonker2000]

Hi, thanks for your interest in the work.

We are currently looking into the code, and developing a more user freindly way to utilize other datasets. If you have other suggestions for formats let us know and we will also implement them.

We will try to complete these changes by the end of next week, however if you urgently need a solution, the easiest way would be to create a variation of the data.py/load_train_val()/data.py/load_train_test() that loads your dataset in a similar format. However, as mentioned, we are currently working on a much simple solution, and will let you know when this solution is implemented.

[T-Almeida]

Hi @JanRodriguez ,

Apologies for the delayed response - we've been focused on meeting several critical deadlines. @richardjonker2000 has implemented an enhanced API in PR #2 that provides better support for external datasets. The implementation details are documented in the README. Please let us know if you need any clarification.
Additionally, we've identified a minor bug (issue #3) in the inference module. Furthermore, we'll also be making some adjustments to inference API itself in the next few days.