What's the difference between normalize_Y=True and exact_feval=False?

[lewisscola]

Hi, I don't know the difference between normalize_Y and exact_feval in GpyOpt. I see the documention like: Is it means that we only have to set one of the parameters? (for example, if we choose normalize_Y=True, we don't have to set exact_feval=False) Could you help me explain it? Thank you!

[astroHaoPeng]

Base on my understanding, they are two different things:

normalize_Y means it will normalize the Y data before doing optimization. But user will always get dimensional output Y. Unless you debug into the detailed optimization processes, you will not interactive with the normalized Y.

Exact_feval means if your evaluation of the objective function f is exact. For example,

• if f_math has a know mathematical formula, usually extact_feval=True unless you contaminate the evaluation of f.
• if f_measure is some a measurement of some physical quantify like temperature, you should gain inexact f_measure and thus extact_feval=False, unless you want to treat the measurement as truth and assume no noise in it.