Open suntzu86 opened 10 years ago
one question: it's possible (probable) that the order in which we add in points_being_sampled
will matter when using GP-dependent heuristics (like kriging). This is irrelevant for constantliar type heuristics. We may want to:
and/or
Right now you can only using constant liar, kriging believer, etc. to solve q,0-EI. No reason why we can't do this with q,p-EI.
The basic algorithm is:
So we can simply first do:
And that provides a way of solving q,p-EI. For standard kriging believer, this will drop the EI at each
points_being_sampled
to 0, allowing the heuristic solve (over q points) to identify new areas of interest.