Github user cestella commented on the issue:
    The issue around the `StellarScientificTest` change is that merging 
t-digests is not precisely the same as adding the same data to one t-digest.  
Whether this is a bug in the t-digest code or whether it's an artifact of the 
theory, I don't know.  Because this test is using low amounts of data, it's 
prone to a bit of jitter it seems and the compression level specified before 
isn't sufficient to ensure that the epsilon chosen (1e-2) is sufficient for 
repeated runs.  I tightened up the compression level to 150 (which is still 
quite sensible) to make sure we can support at least 1e-2 accuracy for normally 
distributed data across 100 runs of 1000 elements.

If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at or file a JIRA ticket
with INFRA.

Reply via email to