Closed TomAugspurger closed 7 years ago
Glad to see activity and interest here. I agree that we can probably be bleeding edge here for a while.
Glad to see activity and interest here.
My (regular, scikit-learn) grid search pipeline that's been running for ~12 hours now because of an expensive step early step in the pipeline, that can be shared across all the grid-search steps, is making this more compelling :)
Ha, indeed. If you happen to have a few computers nearby you could also try scaling out a bit. Not sure if that's worth it in your case though.
Happy to merge when green, or hold off if you have more fixes later. Up to you.
It looks like we've failed to merge here. @jcrist is everything here still good?
Yeah. I'll just merge this now, other changes can come later.
I'm assuming that we don't need to worry too much about backwards compatibility here, though if we do I can put in a compatibility layer. As it stands, the changes to
dklearn.cross_validation
will mean thatdklearn
requires dask 0.12+ (for the changes inrandom_state_data
, cc @jcrist )There are also a couple changes for scikit-learn 0.18 compatibility. I think these changes should be ok for older versions of scikit-learn.
I'm trying out dklearn today, I'll report back with issues / successes :)