Github user akopich commented on the issue: https://github.com/apache/spark/pull/19565 @WeichenXu123, in a case of large dataset this "adjustment" would have infinitesimal effect. (IMO, no adjustment is needed -- the expected number of non-empty docs in the same and does not depend on the order of filter and sample and equals to `docs.size * miniBatchFraction * fractionOfNonEmptyDocs`). So I believe, we all agree that sampling should go before filtering. I'll send a commit shortly.
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org