Query regarding capabilities of SMT

[akki2-4-8]

Hi,

I'm evaluating the capabilities of SMT to perform surrogate based global optimization. The problem consists of optimization of a black-box function having mixed variables (continuous-7 and discrete-3), has multiple objectives (may have linear or non linear constraints also).
To my knowledge:

1. The EGO algorithm presented in SMT is applicable for multi-variable single objective functions and there is another multi-objective EGO algorithm for multi-variable multi-objective functions to perform the infilling using EI strategy?
2. In case there is a way to perform multi-objective surrogate based global optimization of black-box functions, is there a possibility to plot the Pareto frontier for up to 3 objective functions?
3. In case 2. is possible, can SMT handle the mixed variable problems?

Any guidance / suggestions would be of great help.

Thanks & regards,
Akshat

[relf]

Hi, thank you for your interest. As for now, SMT EGO does not handle multi objectives nor constraints but mixed integer handling has been added recently. There is no plotting feature in SMT (which is focused on algorithms).

That being said, you could build your own multi-objective constrained optimizer on top of SMT surrogates and sampling methods. Contributions are welcomed 😉