Github user wzhfy commented on the issue: https://github.com/apache/spark/pull/19406 Yes it's ok if consider it's approximate with relativeError. But the point of this pr is that there could be a further improvement in those special cases. IMHO it's more accurate and intuitive for end users, such as return 1 instead of 2 as 10% percentile among 1 to 10.
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org