Constructing the sample_pareto_frontiers for the qMultiObjectiveMaxValueEntropy acquisition function

[mamunm]

In the botorch.acquisition.multi_objective.max_value_entropy_search.qMultiObjectiveMaxValueEntropy function, it requires a sample_pareto_frontiers callable, which will be used to construct the num_samples sampled Pareto frontier (link: https://botorch.org/api/_modules/botorch/acquisition/multi_objective/max_value_entropy_search.html#qMultiObjectiveMaxValueEntropy). I couldn't find an example of how to construct it. How do we build the sample_pareto_frontiers for a multiobjective optimization problem?

I am thinking of something like this:

def sample_pareto_frontier(X):
    pareto_front = is_non_dominated(X)
    return X[pareto_front]

But it doesn't look right as the output should be num_samples * n * m.

[esantorella]

Hi, I think the following should work, but let me know if you run into issues with it:

from botorch.acquisition.multi_objective.utils import sample_optimal_points

def sample_pareto_frontier(model):
    pareto_sets, pareto_frontiers = sample_optimal_points(
        model, bounds=bounds, num_samples=num_samples, num_points=num_points
    )
    return pareto_fronts

You would need to define bounds, num_samples, and num_points.