Issues with CuPy and the Sparse Implementation

[crumleyc]

Currently implemented is pydata's sparse.COO but it isn't compatable with cupy yet! It is part of their future plans, but currently not.

Other option but not really. Cupy has scipy coo_matrix implemented but it doesn't support 1D vectors.

The S Matrix is a cupy array, so the 1D array must be cupy or compatible with cupy and numpy aka sparse.COO

[crumleyc]

pydata/sparse#198

[crumleyc]

I have tried other unsuccessful attempts:

v = cp.random.random(176) #(176,)
indices = np.argpartition(v,-20)[-20:] #(20,)
col = cp.ones(20) #(20,)

cupyx.scipy.sparse.coo_matrix((v[indices],(indices,col)))

This results in an error saying that all parameters needs to be 1D arrays, even though they are all 1D.

v = cp.random.random(176) #(176,)
indices = np.argpartition(v,-20)[-20:] #(20,)
min = v[indices].min()
v[v < min] = 0
cupyx.scipy.sparse.coo_matrix(v)

I think this method only works when v is already a sparse matrix.

[crumleyc]

I noticed in dask's distributed repository, they had some unit testing where they implemented the cupyx.scipy.sparse.coo_matrix here. I got that to work.

[crumleyc]

u_new = da.dot(S,_V_.result()).compute() is causing an error:

TypeError: 'coo_matrix' object is not subscriptable

Where V.result() returns a dask cupy sparse array.

Looks like it is being looked into cupy/cupy#3178.