Open beckernick opened 3 years ago
@viclafargue and take a look at this one too?
Yes, some operations on Dask arrays do not work when the array is backed by CuPy. This is apparently the case with the unique
operation. I think the best option would be to make these operations work in Dask.
Depending on the compatibility requirements, we could potentially update the internal unique function to use np.empty_like
or NEP-35
For the 0.19 release, we will probably just raise an exception with an informative message when the MNMG KNN Classifier is given a Dask array backed by CuPy.
XGBoost is failing similarly (cross-referencing for additional tracking) https://github.com/dmlc/xgboost/issues/6820
This issue has been labeled inactive-30d
due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d
if there is no activity in the next 60 days.
This issue has been labeled inactive-90d
due to no recent activity in the past 90 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed.
The issue seems to be resolved with:
>>> dask.__version__
'2024.5.1'
>>> cupy.__version__
'13.2.0'
In the 2021-03-26 nightly (and perhaps earlier),
cuml.dask.neighbors.KNeighborsClassifier
fails duringfit
on dask array input if the arrays are backed by CuPy.