Github user mattf commented on the pull request:
https://github.com/apache/spark/pull/2313#issuecomment-54804358
>> What's the behavior if some slaves have numpy but others do not have?
>
> Yeah, this was my original concern. Do both implementations give equal
results, or will this patch lead to non-deteterministic outcomes if tasks
sometimes run with numpy and sometimes without (e.g. while recomputing failed
tasks)?
i had the same concern. i mostly talked myself out of it: first, the
results are by definition non-deterministic, it's a prng afterall; second,
there can already be different implementations of numpy (or better/worse random
sources) on different nodes and we wouldn't be able to detect them.
if you guys agree, i'll check on the model level impl, which may be nicer
in that it could get rid of _use_numpy
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
