Jimmy Lin
Jimmy Lin
hi @pefimov the repro should work exactly as stated in the doc... you might have gotten yourself in a weird inconsistent state. Try removing everything, including anserini/pyserini files in `~/.cache/`,...
Hi @pefimov unfortunately, it doesn't seem like anyone else has been having this issue...
Hi @pefimov Thanks for following up! What a corner case... Hope Anserini is working out for you otherwise?
Noted, thanks for your feedback.
Hi @pefimov - I don't think that's a bug. For all topic files, the base directory gets appended later. The onboarding path asks you do download the data separately, for...
@FarmersWrap correct - I think the normalize here is typically called "min_max_normalization". > Even though rrf doesn't depend on weights, but the rescored number is different and assigned back to...
Unless @lilyjge thinks otherwise, can @FarmersWrap you refactor along the lines above? In terms of breaking ties... in fact I've written a paper about it: https://dl.acm.org/doi/10.1145/3331184.3331339 > The obvious solution...
To clarify, min_max_normalization is preprocessing, the fusions methods are {rrf, interpolation/avg/weights}?
{interpolation, avg, weighted} are the same with different parameterizations... - if you weight with uniform (e.g., 0.5 and 0.5), then it's averaging - if your weights sum up to one...
Yes, for arbitrary weights, you can rescale so everything sums to one. That's a minor detail... Let's see a PR and have something concrete to discuss?