feat: Implement 'prepare_for_training' for text classification datasets

[dcfidalgo]

This PR implements the prepare_for_training method for the DatasetForTextClassification. I modified the first tutorial to show the usage: https://rubrix.readthedocs.io/en/feat-prepare_for_training/tutorials/01-labeling-finetuning.html
I don't think I will manage to implement the prepare_for_training for all tasks before the release.

@dvsrepo @frascuchon For the TokenClassification task, should we rely on spacy utilities to make the conversion from spans to ner tags, or should we provide our own utilities? To me it would seem a bit strange if we required spacy to be installed, to prepare the dataset for training with transformers ... @dvsrepo Do you know of any other library that provides some utilities for converting spans to ner tags?

[codecov]

Codecov Report

Merging #1209 (e4c0bc8) into master (1ba663f) will increase coverage by 0.04%.
The diff coverage is 97.61%.

@@            Coverage Diff             @@
##           master    #1209      +/-   ##
==========================================
+ Coverage   95.23%   95.28%   +0.04%     
==========================================
  Files         124      124              
  Lines        5143     5155      +12     
==========================================
+ Hits         4898     4912      +14     
+ Misses        245      243       -2     

Flag	Coverage Δ
pytest	95.28% <97.61%> (+0.04%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
src/rubrix/client/datasets.py	98.03% <97.61%> (-0.23%)	⬇️
src/rubrix/client/sdk/users/api.py	100.00% <0.00%> (ø)
src/rubrix/client/rubrix_client.py	94.04% <0.00%> (+1.26%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 1ba663f...e4c0bc8. Read the comment docs.

[dvsrepo]

This PR implements the prepare_for_training method for the DatasetForTextClassification. I modified the first tutorial to show the usage: https://rubrix.readthedocs.io/en/feat-prepare_for_training/tutorials/01-labeling-finetuning.html I don't think I will manage to implement the prepare_for_training for all tasks before the release.

@dvsrepo @frascuchon For the TokenClassification task, should we rely on spacy utilities to make the conversion from spans to ner tags, or should we provide our own utilities? To me it would seem a bit strange if we required spacy to be installed, to prepare the dataset for training with transformers ... @dvsrepo Do you know of any other library that provides some utilities for converting spans to ner tags?

I agree we shouldn't rely on spaCy. Let me think of alternatives

[frascuchon]

This PR implements the prepare_for_training method for the DatasetForTextClassification. I modified the first tutorial to show the usage: https://rubrix.readthedocs.io/en/feat-prepare_for_training/tutorials/01-labeling-finetuning.html I don't think I will manage to implement the prepare_for_training for all tasks before the release.
@dvsrepo @frascuchon For the TokenClassification task, should we rely on spacy utilities to make the conversion from spans to ner tags, or should we provide our own utilities? To me it would seem a bit strange if we required spacy to be installed, to prepare the dataset for training with transformers ... @dvsrepo Do you know of any other library that provides some utilities for converting spans to ner tags?

I agree we shouldn't rely on spaCy. Let me think of alternatives

The server part computes tags at token level. You could use for this purpose. If i remember the computed tags are in the IOB format. It could be a begining

[dvsrepo]

This PR implements the prepare_for_training method for the DatasetForTextClassification. I modified the first tutorial to show the usage: https://rubrix.readthedocs.io/en/feat-prepare_for_training/tutorials/01-labeling-finetuning.html I don't think I will manage to implement the prepare_for_training for all tasks before the release.
@dvsrepo @frascuchon For the TokenClassification task, should we rely on spacy utilities to make the conversion from spans to ner tags, or should we provide our own utilities? To me it would seem a bit strange if we required spacy to be installed, to prepare the dataset for training with transformers ... @dvsrepo Do you know of any other library that provides some utilities for converting spans to ner tags?

I agree we shouldn't rely on spaCy. Let me think of alternatives

The server part computes tags at token level. You could use for this purpose. If i remember the computed tags are in the IOB format. It could be a begining

Definitely, that's more than enough!

[dcfidalgo]

The server part computes tags at token level. You could use for this purpose. If i remember the computed tags are in the IOB format. It could be a begining

Nice, totally forgot about this. Hm, should I copy the logic to the client part? Shall we require the user to upload/load the records before calling prepare_for_training? Or is there an endpoint where I can send the records and get back the metrics/iob tags?

[frascuchon]

Not sure about the user flow here. We can discuss tomorrow in a call @dcfidalgo