Conversation
|
I looked at some cuml 24.06 CI runs and it seems like there are a lot of failures. This PR disables query planning in cuml so that CI isn't blocked. CI failure logs |
|
@bdice do you have a link to the 24.06 failures? On a quick look haven't found a PR that has finished jobs in 24.06 |
|
Thanks @bdice! Interestingly, the errors you posted are from the wheels job, from the conda multigpu job intead I see others along the lines of E AttributeError: 'Scalar' object has no attribute 'compute_chunk_sizes'
/opt/conda/envs/test/lib/python3.9/site-packages/dask_expr/_collection.py:452: AttributeErrorfrom here: https://github.com/rapidsai/cuml/actions/runs/8347760672/job/22849832453 |
|
This PR definitely seems to reduce the number of failures. I still see All the failures appear to be from |
Looks like dask-expr doesn't know how to convert from a Dataframe collection to a I'm sure there are many other issues, but that one shouldn't be too bad. |
when query-planning is enabled, implicit conversion is not yet supported from a cudf-backed collection to a dask array. [Some cuml + dask CI failures are related to this limitation](rapidsai/cuml#5815 (comment)). This PR adds basic support for implicit conversion. Authors: - Richard (Rick) Zamora (https://github.com/rjzamora) Approvers: - Lawrence Mitchell (https://github.com/wence-) URL: #15378
|
As an alternative to this PR, #5835 should also unblock CI now. |
|
I'll close this since #5835 is a better approach than temporarily disabling tests. |
This PR disables dask / dask-expr query planning until some bugs can be fixed.
This is needed to unblock cuml 24.06 CI.
xref: rapidsai/cudf#15027
cc: @rjzamora