Optimization of Acquisition Function subject to Parameter Constraints about ax

[ldworkin]

Hi @chakrabarti ! Thanks for reporting this issue. Someone on our team will look into it ASAP. I'll keep you posted!

[lena-kashtelyan]

Hello, @chakrabarti! I'm seeing the exact same behavior with the Service API; if you still have the example that was working for you, it would be very helpful to see it! Service API does not support ParameterConstraint, also, so maybe your constraint was different?

In the meantime, we will investigate this further, since it seems that this is not simply an API issue.

[chakrabarti]

It seems like I can't replicate this exact problem in the Service API. I think perhaps this issue only occurs if the constraint plane is orthogonal to one of the dimensions (so that the vector defining the constraint only has 1 non-zero entry and the constraint is essentially a bound on the variable corresponding to the non-zero entry).

However, I am still curious why this happens in this case (I want to be sure that this doesn't occur for constraints involving multiple variables and even though it seems to work, it seems like there isn't anything fundamentally different between the two cases). Is the constrained optimization guaranteed to work if multiple variables are involved in the constraints (barring an infeasible region or some other issue like that)?