Problem Bayesian Optimisation

[#1]

In an earlier R&D study I sucessfully used a 4 factor mixture DOE (3 mixture factors and one 3-level categorical "Isomer" effect) to find optimal settings for a chemical formulation with three quality attributes Y1, Y2 and Y3.

I used the case in a DOE course and used simulation formulas to compute the responses Y123. Good models are obtained and optimization was straightforward and OK, see "ANALYSIS MIXTURE DOE CHEMICAL FORMULATION" in attachment.

Starting from a 12 run mixture Kowalski model DOE I am struggling with  Bayesian optimization (see annex "BAYESIAN OPTIMIZATION") This keeps failing because of a very unreliable Y2 model? Why does Bayesian recommends "Max Desirability" with this very unreliable Y2 model? Starting from other screening & space filling DOE's  I keep getting the same unreliable Y2 model.

What is the problem? How to solve?

[#2]

On this screen you shared, click on "Check for Updates" (blue text on the left), a window will open and you'll get a link to download the updated version.

Thanks for the info @Florian_Vogt, the release notes about 19.1.x are not detailed enough to see if something has changed in the algorithm/calculations from the BO platform: https://community.jmp.com/t5/JMP-Knowledge-Base/JMP-New-Features-and-Release-Notes/ta-p/575202

[#3 · Accepted solution]

Good Morning @frankderuyck and @Victor_G ,

I can confirm that some of the improvements in the latest release solve the problem:

Before JMP 19.1.4:

With JMP 19.1.4 (current latest release):

Updating to the latest release includes improvements to the platform giving better results (as Victor showed).

Best

Florian

[#4]

Thanks Florian, how to upgrade to latest version?

[#5]

[#6]

Dear Victor and Florian, thanks a lot for you advice, I upgraded to newest JMP version and finally get good results cfr. attachment. Only thing that is not clear is wh after several batches, reason added seems to get stuck on "Maximum Expected Improvement", it does not switch to replication so it seems not to find the optimum, correct? However when I optimize the last batch optimal results are OK, this is not the case after the first Max. improvement. Question is after how many Max. expected improvement to got to optimization, when regression models are OK? What is the most important indicator for Gaussian? 

[#7]

Hi @frankderuyck ,

It's not stuck really...you have an optimization task with 3 responses, where it tries to achieve 3 goals. Since it is still seeing improvement potential, it continues to try to improve. This is indicated in the desirability run chart, as achieved desirability is still improving. When you continue asking for runs, it eventually tries to replicate the current best.

The decision when to stop is tough to generalize - bluntly, you can stop whenever you want, i.e. whenever you feel your goals are met. BayesOpt will try to optimize and verify the results ("Replicate Best Training Run"), depending on the data.

Hope this helps!

Florian

[#8]

Guess the last rows are not the optimal settings, therefore I need to go to model summary and Max Desirability correct? I can see that the profiler has no simulator; when I save prediction formula and go to >graph >profiler it does not work, I get a blank profiler sheet? How to make simulations and check capability with random noise on optimized settings?