Resulting type from multiplication of a sparse and full matrix

General Usage
1

Given a full matrix A and a sparse matrix B, matrix multiplication gives a different resulting type (either full or sparse) depending on the order of multiplication and whether or not a transpose is involved:

julia> A = randn(5,5);

julia> B = sprand(5,5,.1);

julia> typeof(A*B)
Array{Float64,2}

julia> typeof(B*A)
Array{Float64,2}

julia> typeof(A'*B)
SparseMatrixCSC{Float64,Int64}

julia> typeof(A*B')
SparseMatrixCSC{Float64,Int64}

julia> typeof(B*A')
Array{Float64,2}

julia> typeof(B'*A)
Array{Float64,2}

What is the logic behind these resulting types?

1 Like