final variable values and the functions values

[SamirLamin]

I used betorch to solve my problem based on the existing problem integrated in the platform, can you please me show me how to get my final variable values and the functions values

[eytan]

Do you have a code example to reproduce? In our tutorials we plot the final values at each iteration so it should be straightforward. You may wish to try out Ax (http://ax.dev) if you are just using BoTorch as a tool for optimization (for some real world application or research project). BoTorch is intended for users who are developing new Bayesian optimization methods (eg new acquisition functions). Best E

…

On Sat, Feb 18, 2023 at 1:11 PM SamirLamin ***@***.***> wrote: I used betorch to solve my problem based on the existing problem integrated in the platform, can you please me show me how to get my final variable values and the functions values — Reply to this email directly, view it on GitHub <#1690>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAW34NVTO2WY5EFJNW6SWLWYEGELANCNFSM6AAAAAAVAPKQCM> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[SamirLamin]

I create my problem in multi_objective as a class like the BraninCurrin problem and I use it by replacing BraninCurrin by my problem with its characteristic I find the plots but I need to know how the final variable values and the values of the function optimum

[saitcakmak]

Hi @SamirLamin. Assuming you're following a tutorial like https://botorch.org/tutorials/multi_objective_bo, here's how you'd get the points on the Pareto frontier. From that tutorial, suppose that you picked qNEHVI and have access to train_x_qnehvi and train_obj_qnehvi after the BO loop completed. You can use is_non_dominated to find which points are on the Pareto frontier:

from botorch.utils.multi_objective.pareto import is_non_dominated
non_dominated_flag = is_non_dominated(train_obj_qnehvi)  # boolean denoting whether the corresponding objective is non-dominated.
pareto_x = train_x_qnehvi[non_dominated_flag]
pareto_obj = train_obj_qnehvi[non_dominated_flag]

pareto_x gives you the designs and the pareto_obj the corresponding objective values of the designs on the observed Pareto frontier. Note that since this is a multi-objective problem, there isn't a single optimal value but a set of points that are on the observed Pareto frontier.

[SamirLamin]

Ok Thank you so much for your help