On Tuesday, 29 April 2014 19:17:50 UTC+2, Hans W Borchers wrote: > [...] > As Arnold Neumaier once wrote on > > I found CMA-ES quite robust in dimensions up to 50. > (It gets very slow though when the dimension is large.) >
we can make this more specific: with n the search space dimension, the internal effort is O(n^2) per function evaluation (about 1e-8*n^2 seconds for the C implementation with 2GHz processor), and the number of function evaluations scale typically between O(n) and O(n^2). I admit that many years ago I abandoned my R code, because it just seemed to be too slow.
