Fabian Pedregosa

random thought: can we get away with an (unbiased estimator of) the largest eigenvalue? We could get this for example from the power method + russian roulette

I don't know. Now that I think about it its not even clear that for the LMO an unbiased estimator is better than an inexact one (i.e., deterministic but with...

good point

superceded by #336

I approve the version and agree to 4 On Jun 15, 2016 7:59 AM, "Isuru Fernando" [email protected] wrote: > I approve the version and agree to 4. > > —...

Addressed in #211

oops, reopening

oh, got it. It would be nice though if the OptStep was clickable. I don't think it's documented (tried pasting it in the search bar and didn't get anywhere).

Note: ProjectedGradient does have these kwargs documented (https://jaxopt.github.io/stable/_autosummary/jaxopt.ProjectedGradient.html#jaxopt.ProjectedGradient), but I find there's room for improvement also here. For example, the unroll one is "whether to unroll the optimization loop" ,...

test accuracy not great ...