On Tuesday, May 20, 2014 08:34:21 AM Holger Stichnoth wrote: > The only other thing that I discovered when trying out Julia and Optim.jl > is that the optimization is currently considerably slower than Matlab's > fminunc. From the Gist I provided above, are there any potential > performance improvements that I am missing out on?
In addition to the factors that John mentioned, I'll add that it's nontrivial to compare performance fairly. One of the biggest issues is that the algorithms may use different convergence criteria, and thus may have huge differences in the number of iterations. For my problems, Optim's `cg` just so happens to be about 100x faster than fminunc. YMMV. Definitely worth profiling to see where the bottleneck is. --Tim
