Hyperband Is Taking Off, Right?

The method hyperman doesn't just work well in practice on the data sets that we've tried but we've theoretically characterized its behavior, its strengths and its weaknesses. And basically what we're able to show at a very high level is that in optimistic cases it can do significantly better than standard baselines. It's not something where I might work for this problem but it might not work for that problem. There's 18 different hyper parameters of the algorithm itself they need to be set in order to get it to work.