Jackie-Jiang commented on pull request #7076: URL: https://github.com/apache/incubator-pinot/pull/7076#issuecomment-865223114
@tdunning Thanks for the explanation. Yes we randomly generate 100,000 t-digest objects, each constructed from 2 random integers of range [0, 10,000). The difference between star-tree vs non-star-tree is just the merging order of these 100,000 t-digest objects, and eventually we will merge all the t-digest objects into one. Then we compare the result of quantile 0%, 1%, ..., 99%, 100% and check the difference of the results from the 2 merged t-digests. With compression 200, I can still observe difference over 100 from the 2 t-digests (over 1%). While with the old version (3.2), the difference is always smaller than 50 (0.5%). Is this expected? I feel the new version somehow has worse accuracy than the old version -- 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. For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
