Closed 9849842 closed 3 years ago
This is probably due to recent changes in Dask/Distributed. I would suggest trying a newer version of RAPIDS if possible, the current stable version is now 0.17. If for some reason you can't upgrade, what you could try is to downgrade Dask and Distributed packages to versions 2.24.0, as your install is probably picking up a much newer version of those as we don't pin dask-cuda to any specific version of those packages.
Keep in mind that rolling back to an older version of Dask and Distributed packages may require also older versions of different packages too, and I can't ensure that won't be problematic with Dask then. The best solution would indeed be that you upgrade to RAPIDS 0.17.
I have upgraded to RAPIDS 0.17, now I'm getting this:
---------------------------------------------------------------------------
ModuleNotFoundError Traceback (most recent call last)
<ipython-input-1-3a2c34e881e6> in <module>
----> 1 from dask_cuda import LocalCUDACluster
2 from dask.distributed import Client
3 from dask import array as da
4 from dask import dataframe as dd
5 import xgboost as xgb
ModuleNotFoundError: No module named 'dask_cuda'
Your installation is missing dask-cuda, are you installing RAPIDS from conda or some other way? If you're installing from conda the rapids=0.17
metapackage (see https://rapids.ai/start.html) includes dask-cuda, but if you're only installing a selection of packages (e.g., cuDF only), then you need to install the dask-cuda=0.17
package as well.
I am installing from Conda, so I shouldn't be getting that error.
Then please make sure you install the rapids=0.17
metapackage (includes dask-cuda), if you're installing only a subset of that, then install also dask-cuda=0.17
.
I am using the RAPIDS AMI on EC2
What is the AMI number you are using ?
Hmm, actually, I would recommend you follow the instructions here: https://rapids.ai/cloud#AWS-EC2 for launching RAPIDS on EC2
This has gone stale, closing. Please feel free to reopen if there's more to be discussed here.
I am using Rapidsai 0.15 and python 3.7.6
Here is the code I am using to create the client and cluster;
and here is the error after I run that cell:
Here is my GPU setup: