Multidimensional arrays about proximalalgorithms.jl HOT 4 CLOSED

It looks like you're calling the solvers correctly.

Regarding point 1. these are functions that are not exported by ProximalAlgorithms to this package but rather to AbstractOperators and ProximalOperators. So this issue is not related to this package.

Are you saying that having 1. and 2. will get you to all(isapprox.(xp2[13,13,:],1.0;rtol=1e-4)) = true?

from proximalalgorithms.jl.

Hi, and sorry if I wasn't clear. I'm trying to reduce allocations arising from redefining f and A at every iteration. This might be possible by simply using a different approach, i.e., without changing anything in ProximalAlgorithms or AbstractOperators. Or possibly it requires either 1. or 2. Broadcasting would help because f and A could be overwritten in-place.

The issue in the 3D case is that the value xp2[13,13,:] is underestimated possibly due to the computation of gamma with a 3D operator A (I haven't investigated how gamma is computed exactly).

from proximalalgorithms.jl.

I'm not sure that having gamma set to the optimal one will get your condition satisfied. Having optimal gamma might only get you to the optimal solution with less iterations. My impression is that your condition may simply be unfeasible. However, I'm not familiar with your problem.

Concerning the allocations of f and A, this is related to ProximalOperators and AbstractOperator respectively. Just open an issue in the right package and we can discuss about it!

from proximalalgorithms.jl.

Thanks! Will look into the details and open a new issue if the issue persist.

from proximalalgorithms.jl.