MethodError: no method matching Strided.UnsafeStridedView

[LivelyLiz]

I tried to use SparseArrayKit with TensorOperations, but when I attempt to do a contraction, I get this error:

ERROR: LoadError: MethodError: no method matching Strided.UnsafeStridedView(::SparseArray{Float64, 3})
Closest candidates are:
  Strided.UnsafeStridedView(::PermutedDimsArray{T, N, P}) where {T, N, P} at ~/.julia/packages/Strided/pDrSx/src/unsafestridedview.jl:27
  Strided.UnsafeStridedView(::Base.ReshapedArray) at ~/.julia/packages/Strided/pDrSx/src/unsafestridedview.jl:26
  Strided.UnsafeStridedView(::SubArray) at ~/.julia/packages/Strided/pDrSx/src/unsafestridedview.jl:25

Full Stacktrace:

Stacktrace:
 [1] isblascontractable(A::SparseArray{Float64, 3}, p1::Tuple{Int64, Int64}, p2::Tuple{Int64}, C::Symbol)
   @ TensorOperations ~/.julia/packages/TensorOperations/LDxfx/src/implementation/stridedarray.jl:332
 [2] contract!(α::Bool, A::SparseArray{Float64, 3}, CA::Symbol, B::Vector{Float64}, CB::Symbol, β::Bool, C::Matrix{Float64}, oindA::Tuple{Int64, Int64}, cindA::Tuple{Int64}, oindB::Tuple{}, cindB::Tuple{Int64}, indCinoAB::Tuple{Int64, Int64}, syms::Tuple{Symbol, Symbol, Symbol})
   @ TensorOperations ~/.julia/packages/TensorOperations/LDxfx/src/implementation/stridedarray.jl:230
 [3] contract!(α::Bool, A::SparseArray{Float64, 3}, CA::Symbol, B::Vector{Float64}, CB::Symbol, β::Bool, C::Matrix{Float64}, oindA::Tuple{Int64, Int64}, cindA::Tuple{Int64}, oindB::Tuple{}, cindB::Tuple{Int64}, indleft::Tuple{Int64, Int64}, indright::Tuple{}, syms::Tuple{Symbol, Symbol, Symbol})
   @ TensorOperations ~/.julia/packages/TensorOperations/LDxfx/src/implementation/stridedarray.jl:89
 [4] test_contraction(A::SparseArray{Float64, 3}, b::Vector{Float64})
   @ Main ~/Documents/projects/volumeTransform/volumeTransform/scripts/sparsearray_test.jl:14
 [5] top-level scope
   @ ~/Documents/projects/volumeTransform/volumeTransform/scripts/sparsearray_test.jl:24
in expression starting at ~/Documents/projects/volumeTransform/volumeTransform/scripts/sparsearray_test.jl:24

Here is the minimal example:

using SparseArrayKit
using TensorOperations

function make_sparse_tensor()
    A = SparseArray{Float64}(undef, Tuple([10, 10, 10]))
    for i in 1:10
        A[CartesianIndex(i,i,i)] = i
    end
    return A
end

function test_contraction(A, b)
    @tensor begin 
        res[i,j] := A[i, j, k] * b[k]
    end

    return res
end

A = make_sparse_tensor()
b = randn(10)

test_contraction(A, b)

[LivelyLiz]

Ah, apparently the contraction only works if both operands are a SparseArray and a SparseArray - Vector contraction does not work. I guess the issue is solved but it's a bit unfortunate that you cannot mix a sparse and a dense tensor.

[Jutho]

My apologies for the late response; mixing different array types often leads to problems or ambiguities for many things I would think, not only in this package. Aside from the best implementation, simply deciding on the type of the result is already problematic. Different users / use cases might call for different decisions.