Datasets:

emrecan

/

nli_tr_for_simcse

like 0

Hello Albert,

I merged the PR since you work at HuggingFace and I assumed you were fixing the categories all across the hub, but may I ask why? This is actually a semantic similarity/text scoring task, so it doesn't seem right to have it in text classification, the same goes for my other datasets.

Thank you.

I was mentioning this PR: https://huggingface.co/datasets/emrecan/nli_tr_for_simcse/discussions/2#6356cc7d17d306ce1d158907, accidentally wrote a comment to this one.

Hi @emrecan ,

Thanks for your feedback. Feel free to give your opinion in the issue we have opened to centralize all this task renaming: https://github.com/huggingface/datasets/issues/5137

Indeed we are aligning the task names among all open source projects at Hugging Face. You can find the possible values in this file: https://github.com/huggingface/hub-docs/blob/main/js/src/lib/interfaces/Types.ts#L85

We have performed a bulk fix across all datasets on the Hub, just making sensible guesses to map old task names to new ones, but feel free to choose other values from the link above if you think they are more suitable.

For this specific case:

• Before "text-scoring" was a task_category, with task_ids ["semantic-similarity-scoring", "sentiment-scoring"]
• Now all three are task_ids ["text-scoring", "semantic-similarity-scoring", "sentiment-scoring"] under the task_category "text-classification"