Closed maximusunc closed 1 week ago
There are some odd parts in how tie-breaking "pushes down" ranks on tie. The behavior is all in scoring.py, but that code was written by Rosina's team and just ported over to the ARS. I'm open to changing how ties are handled.
Is rank
used anywhere? UI? I was kind of assuming it was just the index of the result, but it sounds more complicated. Anyways, this came about because of the automated tests reports, so I made an issue there to not use rank
in those reports: https://github.com/NCATSTranslator/ARS_Test_Runner/issues/30
I'd have to double check with UI about whether they're using rank, but yeah it's more complicated than an index. I believe it was intended for some tie breaking behavior with weighted mean, though it doesn't work if the tie is the whole set
Gus just confirmed that UI team doesnt use the rank returned from ARS, they calculate rank on their end.
What is the purpose of
rank
in the ARS results, and how is it calculated? I would expect them all to either be unique or to be 1?