fixed duplicate calculation of spearmanr function in metrics wrapper.

[benlipkin]

During _compute, the scipy.stats spearmanr function was called twice, redundantly, once for calculating the score and once for calculating the p-value, under the conditional branch where return_pvalue=True. I adjusted the _compute function to execute the spearmanr function once, store the results tuple in a temporary variable, and then pass the indexed contents to the expected keys of the returned dictionary.

[lhoestq]

Thank you !

I think the same improvement can be applied in the evaluate library, which is the new library we're using to add new metrics :)

https://github.com/huggingface/evaluate/blob/main/metrics/spearmanr/spearmanr.py

[benlipkin]

Great, can open a PR in evaluate as well to optimize this.

Relatedly, I wanted to add a new metric, Kendall Tau (https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.kendalltau.html). If I were to open a PR with the wrapper, description, citation, docstrings, readme, etc. would it make more sense to do that in the datasets or evaluate repo (or both)?

Thanks!

[benlipkin]

PR opened inevaluate library with same minor adjustment: huggingface/evaluate#176

[lhoestq]

If I were to open a PR with the wrapper, description, citation, docstrings, readme, etc. would it make more sense to do that in the datasets or evaluate repo (or both)?

I think you could just add it to evaluate, we're not adding new metrics in this repo anymore