Mathieu Besançon

Given that stdlib PRs are usually slower to get in, we should start working on that

If help is needed there, feel free to ping us

Also, FW only makes sense with constraints / a regularizer. What would be the GalacticOptim way to specify constraints? The only option I saw is for bound constraints

Ok. FW is a bit odd in that regard because it can take arbitrary constraints supported by the MOI solver provided, or some specific form of constraints, as long as...

Thanks, indeed I tracked down some issues in the Julia repo itself. Could it rely on GenericSVD? https://github.com/JuliaLinearAlgebra/GenericSVD.jl/

Edit on this: GenericSVD is deprecated, GenericLinearAlgebra implements all the necessary features I believe?

oh you're right, some macros do take assignment expressions. Would it be sensible not to format around `=` in a macro then?

Running the first example of the tutorial with other types also yields conversion errors. I'm posting the example here since the issue is related, I can move it to a...

You can reformulate a logsumexp as an exponential cone: https://docs.mosek.com/modeling-cookbook/expo.html#log-sum-exp which will work with DiffOpt, would that help?

Ah sorry, hadn't seen the second part. #50 should not be a bother, not sure how well it will work but there are no technical limitations anymore