torch.sparse.sampled_addmm

torch.sparse.sampled_addmm(input, mat1, mat2, *, beta=1., alpha=1., out=None) → Tensor

Performs a matrix multiplication of the dense matrices mat1 and mat2 at the locations specified by the sparsity pattern of input. The matrix input is added to the final result.

Mathematically this performs the following operation:

$$\text{out} = \alpha\ (\text{mat1} \mathbin{@} \text{mat2})*\text{spy}(\text{input}) + \beta\ \text{input}$$

where $\text{spy}(\text{input})$ is the sparsity pattern matrix of input, alpha and beta are the scaling factors. $\text{spy}(\text{input})$ has value 1 at the positions where input has non-zero values, and 0 elsewhere.

Note

input must be a sparse CSR tensor. mat1 and mat2 must be dense tensors.

Parameters

• input (Tensor) – a sparse CSR matrix of shape (m, n) to be added and used to compute the sampled matrix multiplication

• mat1 (Tensor) – a dense matrix of shape (m, k) to be multiplied

• mat2 (Tensor) – a dense matrix of shape (k, n) to be multiplied

Keyword Arguments

• beta (Number, optional) – multiplier for input ($\beta$)

• alpha (Number, optional) – multiplier for $mat1 @ mat2$ ($\alpha$)

• out (Tensor, optional) – output tensor. Ignored if None. Default: None.

Examples:

>>> input = torch.eye(3, device='cuda').to_sparse_csr()
>>> mat1 = torch.randn(3, 5, device='cuda')
>>> mat2 = torch.randn(5, 3, device='cuda')
>>> torch.sparse.sampled_addmm(input, mat1, mat2)
tensor(crow_indices=tensor([0, 1, 2, 3]),
    col_indices=tensor([0, 1, 2]),
    values=tensor([ 0.2847, -0.7805, -0.1900]), device='cuda:0',
    size=(3, 3), nnz=3, layout=torch.sparse_csr)
>>> torch.sparse.sampled_addmm(input, mat1, mat2).to_dense()
tensor([[ 0.2847,  0.0000,  0.0000],
    [ 0.0000, -0.7805,  0.0000],
    [ 0.0000,  0.0000, -0.1900]], device='cuda:0')
>>> torch.sparse.sampled_addmm(input, mat1, mat2, beta=0.5, alpha=0.5)
tensor(crow_indices=tensor([0, 1, 2, 3]),
    col_indices=tensor([0, 1, 2]),
    values=tensor([ 0.1423, -0.3903, -0.0950]), device='cuda:0',
    size=(3, 3), nnz=3, layout=torch.sparse_csr)