Github user WeichenXu123 commented on the issue: https://github.com/apache/spark/pull/19565 Yes, it changed the probability of samples indeed compared with current code. But according to the comments coming from @jkbradley in #18924 , "in order to make **corpusSize**, batchSize, and nonEmptyDocsN all refer to the same filtered corpus", I think @jkbradley 's meaning is to filter out empty docs before sampling, the relationship should be: `batchSize = miniBatchFraction * corpusSize` and `batchSize == nonEmptyDocsN`
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org