Nevermind, I guess Solr computes cosine similarity and not cosine distance, and the returned score is probably the cosine similarity value.
Cheers, frederic -- Frederic Font - ffont.github.io Music Technology Group, UPF - mtg.upf.edu <https://www.upf.edu/web/mtg/> Freesound - freesound.org El ds, 27 gen. 2024 a les 11:57 Frederic Font Corbera <frederic.f...@upf.edu> va escriure: > Hi everyone, > > I successfully added dense vector search to my Solr-based app, but I’d > like to compare the results with other nn-search solutions and for that > reason it would be good to have access to the actual distance values > computer by Solr. This does not seem to be possible and I can only access > the resulting “score”. That score is of course related to the distance > metric but I could not find information about how this relation works. In > fact, cosine distance is, well, a distance metric (big values=dissimilar > items), but the returned score is a “similarity” metric (big values=similar > items). So how does Solr transform the distance metric to a similarity > metric? Is this documented somewhere? > > Thanks a lot! > > frederic > > > -- > Frederic Font - ffont.github.io > Music Technology Group, UPF - mtg.upf.edu <https://www.upf.edu/web/mtg/> > Freesound - freesound.org > >
