Is Hyperband the Right Tool for Hyper Parameter Optimization?

There's a huge amount of upside to it and very, very low risk of it actually not forming reasonably. And in the worst case it's pretty much as good as those baselines. But there are some system subtleties that are not obvious when you're just thinking about the problem of hyper parameter optimization in the abstract. The high level method is certainly, you know, derived from hyperband as a starting point, but there's a bunch of bells and whistles that are very much motivated by practical systems constraints which I think is pretty interesting in and of itself.