If you used the Series and AnalyzedSeries classes the p-value fix should not affect you because those classes use an AnalysisOptions object that has is own defaults and I didn't change those. (Rather, my change aligns default arguments later in the call stack with those set in AnalysisOptions.)
But it's always good to test of course. henrik On Mon, Dec 9, 2024 at 1:05 AM Alexander Sorokoumov < aleksandr.sorokou...@gmail.com> wrote: > Hi Henrik, > > Thank you for preparing these commits! I will test them on Kafka Streams > benchmark results in the next couple of days. I am especially curious about > the p-value fix. > > Best, > Alex > > > > On Sun, Dec 8, 2024 at 12:30 PM Henrik Ingo <hen...@nyrkio.com> wrote: > > > Hi all > > > > I didn't yet receive powers to push the entire hunter repo into > > https://github.com/apache/hunter > > > > ...and since there isn't any code or branches there, I also cannot > create a > > pull request just yet. But... I did spend this Sunday afternoon > preparing a > > branch that is rebased on top of what I expect to become the canonical > > upstream: > > > > https://github.com/nyrkio/hunter/tree/to-asf-upstream2 > > > > Apparently our fork is 6 commits ahead. I will submit separate PRs once > > there is a repo in place to submit PRs against. But you can already > take a > > sneak peek from above. > > > > The two most interesting ones should be: > > - Add to_json() and from_json() serialization methods' > > - Incremental Hunter: Since Datastax introduced an approach where we > only > > consider the w(indown length) closest points, it turns out that the next > > logical step is an optimization: since the common case is that new test > > results are appended to the end, we can limit the recompute to the 1-2 > > last windows at the tail end. Everything before that point is guaranteed > to > > not have changed, because the newly added test result is not inside the > > window that is considered as input for the algorithm. > > > > Other changes are fixes or an additional test/benchmark. Some are quite > > interesting stlil. Especially I can tell with confidence that Piotr never > > thought of using a p-value of 0.1 or higher :-) (You can look at the > patch > > yourself to find out.) > > > > henrik > > >
