search

andim / noisyopt

Python library for optimizing noisy functions.
http://noisyopt.readthedocs.io/
MIT License
87 stars 15 forks source link

Implement further algorithms for noisy optimization #2

Open andim opened 8 years ago

andim commented 8 years ago

There are a number of algorithms in the literature for the optimization of noisy functions that might make for a good addition to the package. I'm thinking especially about differential evolution based algorithms, on which there seems to be quite some literature (see e.g. http://dx.doi.org/10.1109/CEC.2006.1688534). Any input from practitioners or Pull Requests are welcome.

krishnapitike commented 5 years ago

Is there a way to optimize a function which returns a noisy-vector rather than a noisy-scalar? Thanks in advance!

andim commented 5 years ago

A simple way is to turn the noisy vector into a noisy scalar by combining the various objectives in some weighted manner. There is currently no functionality to find Pareto optimal solutions to multiobjective optimization problems, but if you have ideas in that direction that would be more than welcome.

thomasahle commented 4 years ago

I wrote a baysian optimization based optimizer for noisy functions here https://github.com/thomasahle/noisy-bayesian-optimization . The goal is to approximate what they did for AlphaGo https://arxiv.org/abs/1812.06855 .

Another interesting algorithm is CLOP which is used by many chess engines https://www.remi-coulom.fr/CLOP/