Mattias Fält

Results 60 comments of Mattias Fält

I think this is a very interesting idea, but I think we have to think it through really carefully. I think there are a couple of trade-offs regarding how much...

Didn't mean to push the benchmark, will remove those. This PR solves part of https://github.com/kul-forbes/ProximalOperators.jl/issues/52 (again)

> > Could you also update `IndPolyhedral` as part of this, and add the option for using this variant? This should already be in the PR, or did you mean...

> I’m open to changing the default, let’s maybe look into a more extensive comparison for this type of projections, including ill-conditioned problems where some solvers may have a hard...

The benchmarks for random problems with bounds either as `lb

Just took a quick look, looks like a reasonable example, not sure what we currently have. However, this line `z = @view x[:]` doesn't make much sense. It creates a...

I think it may also be dependent on the order the tests on the different platforms finishes. They seem to differ by the amount that is reported. I haven't verified...

SuiteSparse solve2 should support `Float64, Complex{Float32}` and "zomplex" `Complex{Float64}` but I didn't need them so I only implemented CholmodSolve2.jl for Float64. It should be a relatively trivial to update the...

> That's a good idea, even before any multi-threading is introduced: after a quick benchmark, I observed an increase in time of ~10% for NormL2 and ~270% (!!!) for IndBox,...

I just submitted a paper for a QP solver. It is supposed to be really efficent projecting on Ax=b, Cx