embeddings-benchmark/mteb

Mismatch of the category of AmazonPolarityClassification

twadada opened this issue · 2 comments

In the current ver., AmazonPolarityClassification is registered as 's2s' but it was 'p2p' in the previous ver., and the paper also says it's p2p. I looked at the actual input text, and I guess p2p is the correct one?

Added a fix in #1220 - @Muennighoff just adding you here in case there is something I missed, but I do agree with @twadada that it should probably be s2s

Regardless: "avg_character_length" available in descriptive_stats is probably a better metric for many use-cases.