Max Balandat

So if I read this correctly, "parameter1" in this setting isn't really a tunable parameter but instead an observed feature? That is, its value can help explain the behavior of...

> One idea I'm contemplating, which is only feasible due to these extended labeling periods, involves making predictions with the already trained model. I would fix Parameter1 to its known...

> Still don't understand what the Runner is doing though haha The purpose of the Runner in general is to abstract away how exactly you'd evaluate a Trial provide a...

@cheeseheist thanks for this keen observation, this is very helpful. Seems like we were transforming the search space in-place when calling `evaluate_acquisition_function()`, which then leads to an issue during the...

So this sounds like an active learning type problem where your goal is to find the best model across the whole design space ("best" in some yet-to-be specified sense)? This...

> So in this case can the surrogate model be any model I choose? Or are there any limitations? Or is the best choice here a GP. So in general...

yup, that makes sense to me!

Hmm I think you may be using `qNegIntegratedPosteriorVariance` incorrectly here. Note that the [`mc_points`](https://github.com/pytorch/botorch/blob/main/botorch/acquisition/active_learning.py#L65-L68) arguments takes a (typically relatively dense) set of points across the domain that are used for...

> I went ahead and added a subset of the data mc_points = 500. So what it does is select randomly 500 points from the candidates space and uses it...

So here you're effectively optimizing the MES acquisition function on a finite set of random candidate points. That's not wrong per se, but one of the big benefits of the...