Question: Role of MCSampler in qExpectedHypervolumeImprovement

[1bing2]

Hello，everyone:
I am new here. I have a question to consult.When I read the botorch tutorials, the function called qExpectedHypervolumeImprovement
have a MCSampler ,how can l understand this MCsampler?
or Can you explain the usage of this or give some information about it?

[Balandat]

qExpectedHypervolumeImprovement (and more generally other Monte Carlo acquisition functions in BoTorch) uses Monte Carlo sampling from the surrogate model's posterior in order to compute the acquisition function. For a general introduction to this in BoTorch see sections 3+4 of https://proceedings.neurips.cc/paper/2020/file/f5b1b89d98b7286673128a5fb112cb9a-Paper.pdf, for the specifics about how it's used in qExpectedHypervolumeImprovement, see https://proceedings.neurips.cc/paper/2020/file/6fec24eac8f18ed793f5eaad3dd7977c-Paper.pdf.

The MCSampler is essentially an abstraction that makes it convenient to draw samples from model posteriors while potentially fixing the base samples for Sample Average Approximation (SAA) - see the linked BoTorch paper for more details.