Lorenzo Stella
Lorenzo Stella
@Shuvomoy is \Sigma symmetric positive definite?
This could this be an instance of the more general case g(x) = f(Ax), for convex, proximable f, I guess.
PR #54 contains an implementation of this (no warm-start yet).
Hey @mfalt: all the `gradient!` operations are there, just lazily implemented :-p Jokes apart, you can surely implement the missing ones if you want, and I also completely agree with...
No I haven't opened it in a while, I'll go look at it asap
True, I’m wondering what would that yield in case of nonsmooth functions. It is worth checking for sure.
There are some problems which I found using ReverseDiff.jl, in particular using `ReverseDiff.gradient!` for computing gradients. The problems already show up with very simple functions: take the `LeastSquaresDirect` function, whose...
With a significantly improved Julia AD landscape nowadays, I’m wondering whether it makes sense to keep all these implementations around. For many functions (and proximal operators) the existing AD systems...
Very interesting! I think it would be nice to replace the current implementation. For reference, some test cases on this projection can be found here: https://github.com/JuliaFirstOrder/ProximalOperators.jl/blob/7889da3a39e65b9e601b260a37ccde322914af79/test/test_results.jl#L196
This could be the occasion to first add benchmarks for the projection on the exponential cone, so to “touch with hand” the performance improvement https://github.com/JuliaFirstOrder/ProximalOperators.jl/blob/master/benchmark/benchmarks.jl