tdunning commented on issue #409: URL: https://github.com/apache/datasketches-cpp/issues/409#issuecomment-1886402751
I would recommend a higher default compression factor than 100. But otherwise, your results look good. The reason for the increased accuracy in some cases is that by merging you are effectively keeping a lot of extra centroids in all of the t-digests that you are merging together. This was the motivation behind the way that the merging digest doesn't fully compact the centroids in the buffer until persisting the digest. This has a cost in terms of buffer space (and thus requires a few percent more merges) but wins in accuracy. On Wed, Jan 10, 2024 at 10:09 PM Alexander Saydakov < ***@***.***> wrote: > Now 32-way merge only > TDigest.100.rank.error.after.32-way.merge.0.99.uniform.random.input.png.png > (view on web) > <https://github.com/apache/datasketches-cpp/assets/13126686/4d6489b3-119e-46fe-a529-da2f3ab1aef5> > > — > Reply to this email directly, view it on GitHub > <https://github.com/apache/datasketches-cpp/issues/409#issuecomment-1886353997>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/AAB5E6XIHDSMQSR5ZPQC2TTYN5625AVCNFSM6AAAAABANDLXNOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQOBWGM2TGOJZG4> > . > You are receiving this because you commented.Message ID: > ***@***.***> > -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
