cuspa.tl.CSRMatrix

class cuspa.tl.CSRMatrix(

indptr: Any,

indices: Any,

data: Any,

shape: tuple[int, int],

)

CSR matrix backed by CuPy, CUDA PyTorch, or NumPy arrays.

indptr and indices share one dtype: int32, or int64 when the matrix has more than 2**31 - 1 non-zeros. data is always int32.

Attributes

indptr
indices
data
shape
nnz

Methods

to_cupy_csr([dtype])	Wrap as cupyx.scipy.sparse.csr_matrix.
to_torch_sparse([dtype])	Wrap as a torch.sparse_csr_tensor.

CSRMatrix.indptr: Any

CSRMatrix.indices: Any

CSRMatrix.data: Any

CSRMatrix.shape: tuple[int, int]

CSRMatrix.nnz

CSRMatrix.to_cupy_csr(dtype: Any = 'float32') → Any

Wrap as cupyx.scipy.sparse.csr_matrix.

cupyx.scipy.sparse only supports float and complex data. Counts are cast to dtype, which defaults to float32.

cupyx.scipy.sparse also only supports int32 indices, so this raises OverflowError for a matrix with more than 2**31 - 1 non-zeros. Use to_torch_sparse() or the arrays directly instead.

CSRMatrix.to_torch_sparse(dtype: Any = None) → Any

Wrap as a torch.sparse_csr_tensor.