search

ppgaluzio / MOBOpt

Multi-objective Bayesian optimization
MIT License
81 stars 23 forks source link

Normalization #7

Open VahidMostofi opened 4 years ago

VahidMostofi commented 4 years ago

Hi, Thanks for the great Library. I was wondering do I need to normalize different objective functions? In case they are in different ranges. For example, f1, f2 are in [0,20000] and f3 is in [0,30]

Thank you in advance

ppgaluzio commented 4 years ago

Hello, In principle, there is no need to normalize the objectives. Consider for example the case in which your objectives are very expensive black boxes, you probably wouldn't even know what the normalizations factors should be.

However, if you do know the ranges of the functions, if anything, I would only expect an improvement on convergence rate or precision for the method. It is important to say that I haven't performed any tests on this matter, so this is mostly out of intuition.

It you eventually use it with and without normalization, I would be curious to know if you observed any improvement.