Repository navigation
[ENH] testing ray parallelization but avoiding long test runtimes and hangs #163
Description
Activity
Thanks for the context! 👋
I reviewed sktime/sktime#8149 and the related discussion around Ray reliability in CI. Skipping Ray-dependent tests for now makes sense given the instability and runtime concerns.
I’ll keep an eye on the progress upstream, and once a stable fix or pattern emerges, I can help apply the same approach here (either in the current Hyperactive setup or after the scikit-base refactor, depending on timing).
In the meantime, if there's any smaller isolated task—like marking tests with expected failures, adding better CI guards, or improving documentation around Ray usage—I’m happy to pick that up.
Reacted by Franz Király
Currently, we have adopted the
sktimeparallelization backend dispatcher, inheriting the issues withrayand testingray, e.g., long runtimes, sporadic crashes in the CI - parallel issue insktimehere: sktime/sktime#8149We should monitor
sktime#8149 and apply the same fix tohyperactive(orscikit-baseafter a potential refactor). For now, we have decided to skip therayrelated testing.