Max Balandat

> Correct me if i am wrong here, the candidate_set in the notebook, would be my X_candidates that i am looking to evaluate..? Not quite. The `qMaxValueEntropy` acquisition function, similar...

Indeed, you're evaluating the acquisition function on a discrete set here using `evaluate_acquisition_function`, so you can just pick the best and not use `optimize_acqf`.

> Ok so I think I found the issue. For some reason when you call model_bridge.evaluate_acquisition_function you must use the acq_options argument when using the qNIPV acquisition function, even if...

> @Balandat would it be easier to scale to `[0, 1]^d` outside of Ax using sklearn's [MinMaxScaler](https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.MinMaxScaler.html)? Or, assuming `UnitX` is the only transform at play, would you recommend using...

Hmm yeah that seems like it should work! I guess since `UnitX` transforms to the unit cube it should be enough to just generate the points directly without Ax involvement:...

Yes this is indeed a super interesting application. I second Eytan's point about 3 observations being too few to compute meaningful estimates of the standard error. So if it's the...

> If the total of evaluation budget beyond the initial point is 15 what is the best strategy : 3*5 parallel , 10 parallel +2 parallel or once single batch...

> When I run several time my code on the same initial data, the algorithm do not give always the same value as "next experiement", is it normal ? And...

> After a discussions with @sgbaird, I was wondering if is better to let the system infer the noise or to run the code with all data available (3 repetitions...

> For the first step I have 20 points, with 3 runs (plot bellow), does make any sense to train the model by using all point so 60 points or...