Jacobian dimensions are wrong about nlsolversbase.jl HOT 8 CLOSED

Also it doesn't consider the complex case.

from nlsolversbase.jl.

You're correct if you consider a larger class of problems than we do. Is it to accommodate least squares problems, or what do you have in mind? I actually thought of changing it exactly to make it work with levenberg-marquardt in lsqfit

from nlsolversbase.jl.

I was merely entertaining the possibility with no use case in mind, just for sake of generality since you already have F input, and there seems to be a direction to accommodate complex number optimization.

from nlsolversbase.jl.

Ah okay, yeah we can change it.

from nlsolversbase.jl.

I am also a bit wary of initializing derivatives or Jacobians with 0, since any algorithm bug which checks for convergence criteria first thing in a while loop for example may assume the algorithm has converged. I think this can be one less thing to keep in mind if it is initialized to NaNs or something.

from nlsolversbase.jl.

not anymore

from nlsolversbase.jl.

Awesome!

from nlsolversbase.jl.

:)

from nlsolversbase.jl.