Github user mengxr commented on the issue: https://github.com/apache/spark/pull/22790 This shouldn't block 2.4.0 release. Based on the code, it doesn't introduce regression to existing features (just using V1 format and ignore trainingCost and distanceMeasure). Correctness issue occurs only when someone uses a non-default distanceMeasure and then save/load. Could someone help confirm? If current vote passes, we can list it as an known issue in the release notes and fix it in 2.4.1. If other blockers show up, we fix it before RC5. Btw, this PR needs a regression test in order to merge.
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org