Question about constraints

[FelixNeutatz]

Hi everybody,

I have a question about this line https://github.com/EmuKit/emukit/blob/9f6763f8a0b689b0cd05b64dbdbe27ba6d7953c3/integration_tests/emukit/bayesian_optimization/test_unknown_constraints_bayesian_optimization.py#L14

If I have a constraint x <= 10. Then, I should implement the function:

def fc(x):
    return float(10.0-x)

Is this correct? I did not find any documentation on whether the constraint value is maximized or minimized.

Thank you for your help.

Best regards, Felix

[apaleyes]

Hello! This particular loop is built with Probability Of Feasibility acquisition, which assumes c(x) <= 0, see here: https://github.com/EmuKit/emukit/blob/main/emukit/bayesian_optimization/acquisitions/probability_of_feasibility.py#L18

I do agree though the documentation shall be much clearer, we'll look into improving the docs of UnknownConstraintBO classes. Thanks to bringing this up!