Given that this release is probably still 2 weeks from landing, I don't think that waiting on a spark-tensorflow-connector release with TF 1.12 in mid-October is a big deal. Users can use the library with Spark 2.3.x for a week or two before upgrading, if that's the case. I think this kind of bug fix is appropriate for a minor release, while I could see trying to work around to keep the buggy behavior in a maintenance release. On Mon, Oct 1, 2018 at 12:11 PM Xiangrui Meng <m...@databricks.com> wrote:
> > IMHO, the use case (spark-tensorflow-connector) is very important. But > whether we need to fix it in 2.4 branch depends on the release timeline. > See my comment in the JIRA: > https://issues.apache.org/jira/browse/SPARK-25378 > >