18267 1x1 NaN matrix eigenvalue

[mcognetta]

Added a check when computing the eigenvalue of a matrix for if the matrix is a 1x1 matrix with NaN elements. Previously this case incorrectly returned a value (not NaN) while the 2x2 and larger dimensional cases threw an error.

This is my first contribution to Julia, so any help or feedback is greatly appreciated.

#18267

[andreasnoack]

I think the problem is the issymmetric function. It shouldn't return true for the 1x1 case when the value is NaN. I think the right fix shoud be to also test the diagonal in https://github.com/JuliaLang/julia/blob/master/base/linalg/generic.jl#L702

[KristofferC]

How about issymmetric(NaN)?

[mcognetta]

I am not so sure about the 1x1 issymmetrical case. It is the "same" NaN when transposed so maybe it isn't incorrect for it to be considered symmetrical.

Anyway, I can redo it to be checked in the issymmetric function if that is deemed the better solution (obviously you guys know better than me).

[andreasnoack]

Symmetry is A[i,j] == A[j,i] for all i,j so we just inherit the behavior. The definition for Number should be changed to

issymmetric(x::Number) = x == x

[mcognetta]

I updated it. I did not realize that Julia didn't evaluate NaN == NaN in any case.

I am going to submit a new pr for the issymmetric(x::Number) = x == x since that probably shouldn't be looped in here.

By the way, I know about methods() and applicable(), but is there a way to tell, given a multiple dispatch method name and a possible input, which method specifically would be called (the line number and the parameter types/function signature)? When I was trying to determine which issymmetric and eigfact! was being called, I had to resort to putting print statements in every candidate and rebuilding, which was inefficient.

[simonster]

By the way, I know about methods() and applicable(), but is there a way to tell, given a multiple dispatch method name and a possible input, which method specifically would be called (the line number and the parameter types/function signature)? When I was trying to determine which issymmetric and eigfact! was being called, I had to resort to putting print statements in every candidate and rebuilding, which was inefficient.

Try @which f(x) (or @edit f(x) if you want to open it in your editor)

[tkelman]

probably rand(eltya, 3, 3) rand(typeof(eltya), 3, 3) to avoid repeating this for Float64 3 times?

[mcognetta]

Yes, I did not consider this behavior. Thanks.

[tkelman]

missing a close paren

[andreasnoack]

I don't think that this is sufficient although it will fix the eig issue. For larger matrices, we'd have to check that whole diagonal so instead the loop ranges below should be made one larger to include the diagonal.

[andreasnoack]

@mcognetta It would be great to get this merged soon. Do you have time to finish up the PR?

[mcognetta]

Sorry, been a little busy. Ill do it now.

[andreasnoack]

Thanks for fixing this.

[mcognetta]

Thanks for the help.