Following the issue in MKL package wrong results - #3 by Enlil50 it seems that wrong result were brought upon little (non)zeros of order 10^-16 roaming around the matrix. Whereas for dense matrices I know 2 methods for ensuring symmetricity:
- A_sym = 0.5*(A + A’)
- A_sym = Symmetric(A) #this is just a view of the upper triangular section
For sparse matrix I can use just the first one. It seems that it doesn’t change much. just to approach the issue in a different way: is there something which suppresses numbers under a certain tolerance or threshold in a given sparse matrix ,matrix or array? I would like to try it first.
P.S: i expect that in sparse matrix the zeros are treated just like sparse matrix zeros: they should not be considered in linear algebra operations and such. I don’t need to reallocate the vectors, but it could even give me a little boost.