As stated in the question I need a way to optimise a costly blackbox function which I cannot compute too many times. Can you point me to some ways of doing this? While somehow closing in on the global minima would be best, it would also be okay if we can just find the local minima near the given point.
Asked
Active
Viewed 45 times
0
-
The linked question does not answer this problem specifically. Optimizing over _costly_ evaluations is very different than optimization of simple functions. The second is what is usually implied, and what is asked+answered on the (not) duplicate Q&A. For expensive spaces you should look into the methods used for hyperparameter optimization, for example in [hyperopt](https://github.com/hyperopt/hyperopt). – mcskinner Apr 17 '20 at 21:00
-
Thanks for the suggestion @mcskinner. But does hyperopt somehow use the the initial point I have? I was looking at a similar library rbfopt but couldn't make it use my initial point effectively. – gst1502 Apr 18 '20 at 02:10
-
Also do I have to ask this question again? – gst1502 Apr 18 '20 at 02:12
-
I suppose so. You can point out in the question that you have searched existing posts and this is different. – mcskinner Apr 18 '20 at 04:08