Patrick Hoefler
Patrick Hoefler
cc @jrbourbeau maybe for the release but it's not urgent. Please don't add to the list in config.py, since this raises an error. The option was pretty much useless before,...
feature went away a few months ago and nobody complained, so no need to deprecate now
I want to bring this up again. Yes the expectation in pandas is that users don't set this explicitly. This is one of the reasons why adding it to the...
Thanks for your report. parquet automatically reorders the number of partitions, round tripping through arrays is currently not what I would advise to do. You can set ``optimize=False`` in the...
Thanks for the context, that helps me understand this better. not calling persist would also fix stuff btw, the optimizer will then take both branches into account and will return...
This seems to work on main and the newest release
Thanks for your report. Dask sets the variables OMP_NUM_THREADS, MKL_NUM_THREADS and OPENBLAS_NUM_THREADS to 1 when a client is created. Can you try unsetting the 3 to see if that speeds...
Can you provide something this is reproducible and give us the dask version that you are running things with?
Could you create an example that doesn't rely on external files? I.e. something that we can copy paste. See https://matthewrocklin.com/minimal-bug-reports for a more detailed description
I've added a HLG layer for take here: https://github.com/dask/dask/pull/11256 The task graph that we upload to the scheduler is smaller now, but that's about the only improvement. the scheduler is...