Github user akopich commented on the issue: https://github.com/apache/spark/pull/19565 I am sure that caching may by avoided here. Hence, it should not be used. @srowen, maybe I don't get something, but I'm afraid, that currently lineage for a single mini-batch submission looks like this `docs.filter(nonEmpty).sample(minibatchFraction).treeAggregate(...)`. And I'm afraid that for each mini-batch `filter` will be performed all over again. But if we have smth like `docs.sample(minibatchFraction).filter(nonEmpty).treeAggregate(...)`, this will be avoided. And no caching is needed.
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org