search

rapidsai / deployment

RAPIDS Deployment Documentation
https://docs.rapids.ai/deployment/stable/
9 stars 28 forks source link

RAPIDS + Dask-ML sample notebooks #301

Open xyz00000 opened 9 months ago

xyz00000 commented 9 months ago

getting this:

Exception: "TypeError('Implicit conversion to a host NumPy array via array is not allowed, To explicitly construct a GPU matrix, consider using .to_cupy()\nTo explicitly construct a host matrix, consider using .to_numpy().')"

after running cell: mode = "gpu-random" model_rf = RandomForestClassifier()

Any ideas?

mode = "gpu-random" with timed("XGB-" + mode): & mode = "gpu-grid" with timed("XGB-" + mode):

works great.

jacobtomlinson commented 9 months ago

Could you let us know which notebook you're having trouble with and which versions of RAPIDS you're running? Thanks :)

xyz00000 commented 9 months ago

thanks @jacobtomlinson (jacob), rapids 23.10 (cuda11_py310_231011_gd3958fe_0 rapidsai) and jupyter notebook (pip install jupyterlab 4.0.9), best steve

https://docs.rapids.ai/deployment/stable/examples/xgboost-randomforest-gpu-hpo-dask/notebook/

jacobtomlinson commented 9 months ago

Ok great, I'm going to transfer this issue over to our deployment docs repo where that notebook lives.

We are about to do some testing of these notebooks ahead of the 23.12 release, so I'll ensure this one is high priority.

skirui-source commented 9 months ago

I was able to run through the notebook and also seeing the same error :

image

xyz00000 commented 9 months ago

@skirui-source im 7 years self taught python but i think it's something to do with how dask-ml (2023.3.24) feeds the test data (maybe attempting an additional conversion) into the random forest model.

@jacobtomlinson Would be nice to get a feature_importances (sklearn) through the gpu too?