Custom Desirability in Bayesian Optimization

[#1]

I am trying to do multi response Bayesian optimization in JMP Pro - Apologies if this is a stupid question but, how do I set custom desirabilites for each response and have them carry through multiple iterations? When I use "set desirabilities" it is not clear these are carried through to the run / batch  selection process OR to any further BO iterations.

[#2 · Accepted solution]

Hi @OliverPickburn,

Once you have set or modified the desirabilities of your responses, you can use the option "Save Desirabilities" in the red triangle next to Augmented Prediction Profiler (or next to any Profiler):

• Original situation with desirabilities :
  
  The run added using the default "Max Expected Improvement" has these factor values:
  

• If I change the desirability  of MODULUS to minimize it, then Save Desirabilities, and relaunch the platform with the same settings:  
  

  the default added run has changed:

  

So to answer completely your question, changing desirabilities will not change the candidate set created to rank the points to test in next iterations, but it will change the ranking and selection of points as it will be linked to acquisition function and desirability functions.

Like for DoEs, the Bayesian Optimization platform do use and take advantage of several metadata/information from the column properties (coding, design role, response limits, etc...) or table script (constraints...).

Hope my answer is clear,

[#3]

Clear Victor and Thank you for the response.

It feels slightly "roundabout" given the focus in BayesOpt on optimization, but it works and I can live with that. I will just need to remember to set the response limits before running the platform in future,

Thanks again

Oliver