Sparse_sampled_addmm

Usage

torch_sparse_sampled_addmm(self, mat1, mat2, beta = 1L, alpha = 1L)

Arguments

self: (Tensor) a sparse CSR matrix of size (m, n) to be added and used to compute the sampled matrix multiplication
mat1: (Tensor) a dense matrix of size (m, k) to be multiplied
mat2: (Tensor) a dense matrix of size (k, n) to be multiplied
beta: (Number, optional) multiplier for self ($\beta$)
alpha: (Number, optional) multiplier for $mat1 @ mat2$ ($\alpha$)

sparse_sampled_addmm(self, mat1, mat2, *, beta=1, alpha=1) -> Tensor

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

Mathematically, this performs the following operation:

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

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

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

Examples

if (torch_is_installed()) {

# Create a sparse CSR mask from a COO tensor
i <- torch_tensor(matrix(c(1, 2, 2, 1, 2, 3), nrow = 2, byrow = TRUE),
  dtype = torch_long())
v <- torch_tensor(c(1, 2, 3), dtype = torch_float32())
sparse_mask <- torch_sparse_coo_tensor(i, v, c(2, 3))$to_sparse_csr()

mat1 <- torch_randn(c(2, 3))
mat2 <- torch_randn(c(3, 3))

result <- torch_sparse_sampled_addmm(sparse_mask, mat1, mat2)
result$to_dense()

}
#> torch_tensor
#> -0.0804  0.0000  0.0000
#>  0.0000  3.8992  5.0012
#> [ CPUFloatType{2,3} ]