[BUG] model save problem in cuml

[skyqwe123]

@msarahan @seberg @benfred @mattf Thanks for your good job! When i use cuml, the model can not be saved like sklearn.

Describe the bug
Models cannot be saved like sklearn.

if we use sklearn, it works:
import torch
from sklearn import random_projection
model = random_projection.SparseRandomProjection(eps=0.1)
torch.save(model, 'model.pkl')

in cuml it not work and give following error:
import cuml
model = cuml.random_projection.SparseRandomProjection(eps=0.1)
torch.save(model, 'model.pkl')

TypeError: self.rand_matD,self.rand_matS cannot be converted to a Python object for pickling

[dantegd]

Thanks for the issue @skyqwe123, I haven't used torch.save before with sklearn estimators, seems like it relies in pickle so this must be exposing a bug in our pickling of Spars Random Projection, so this is something we will work in but don't have an ETA currently.