cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
Try the Materials Informatics Toolkit, which is designed to easily handle SMILES data. This and other helpful add-ins are available in the JMP® Marketplace
Choose Language Hide Translation Bar
Jeff2010
Level I

Question about Maximize Desirability in Prediction Profiler

After maximizing desirability in prediction profiler, the setting of the most significant and dominant factor is at its high level, what should I do next step?

 

Option-1: I was thinking about augment the design by expanding the range of the most significant factor, but my colleague does not agree with that for no reason.

Option-2: I was also thinking about using simulation, but it does not make sense for me, because the optimal setting is high level, unless I extrapolate the prediction model.

Option-3: Just run a confirmation run to verify the optimal setting obtained from prediction profiler. 

 

I very appreciate any comment about it or share your experience. Thanks!

 

3 REPLIES 3
Phil_Kay
Staff

Re: Question about Maximize Desirability in Prediction Profiler

Hi,
What you should do next depends very much on your objectives. It sounds like you have run an experiment and you have gained some useful learning. Why did you do the experiment? What did you set out to achieve? What questions did you have before running the experiment? Have you now got some good answers to those questions? Do you have other questions that you need to answer?
Thanks,
Phil
statman
Super User

Re: Question about Maximize Desirability in Prediction Profiler

Adding to Phil's excellent comments/questions...I'll guess you ran an experiment. Were the factors set at 2-levels?  Is the significant factor continuous? Will that one factor allow you to achieve the desired target value for the response?  If not, certainly some exploration in the direction of + is worthwhile.  Have you replicated the experiment?  Do you know if the significant factor effect is consistent over changing noise? Are you concerned with curvature?

 

My advice is when designing experiments predict ALL possible outcomes and what you would do for each outcome. Example: If factor A is significant  I will..., If A is not significant I will...If I run the experiment and don't create any variation I will... If I run the experiment and create a significant amount of variation and that variation is unassignable to any factor I will...

"All models are wrong, some are useful" G.E.P. Box
statman
Super User

Re: Question about Maximize Desirability in Prediction Profiler

I'll add that simulation, you realize, already has an algorithm which is how you get an output when putting values for inputs. I'm not sure how the simulation was created? I'm not sure if it is representative of your situation? I'm not sure how noise is introduced into the simulation? I'd be leery of simulation...if a factor you are investigating is not in the algorithm, it will show as insignificant. That may not mean the factor has no effect in real life.
"All models are wrong, some are useful" G.E.P. Box