Matthew Rocklin

I encourage you to prototype this, perhaps with `dask.delayed`. http://dask.readthedocs.io/en/latest/delayed-collections.html

Can you do me a favor and try this from git master? On Mon, Aug 22, 2016 at 8:05 PM, dirkbike [email protected] wrote: > I was originally thinking of doing...

Then perhaps you're right that your dict-of-dataframes idea would suit better

Eventually yes, it would be nice for DataFrame to support multiindices. It's non-trivial to change all functions within dask.dataframe to support this. I budget this task at somewhere between a...

Yes, that seems like a reasonable synopsis. We would choose some depth of the multi-index along with to partition. For example we might partition along the second or third step...

Sounds very cool. I encourage you to explore that further.

I definitely agree @vss888 if this is something that you'd like to contribute that would be very welcome!

I don't personally have a ballpark estimate, no. Others might though. On Fri, Jun 28, 2019 at 11:33 PM Marissa wrote: > Hi there. I'm trying to get a handle...

@rjzamora and the RAPIDS folks have been playing with multi-column indexing I think One of the nice things about approximate quantiles is that they're decently robust to sampling. I imagine...

> At the moment the plan for 6028 is to wait a bit until dask and distributed are more stable in other places. FWIW I'd say that code-wise stability has...