How can i add design constraints to modelling

[#1]

Hello JMP Community,

I have a custom DoE with a linear constraint (Disallowed: factorX > factorY). However I can not add this contraint when modelling my data. This results in predictions including factorX > factorY.

Apparently the constraints can be modified in the modelling dialog as seen in picture 1. However I do not have this option (see picture 2).

Any tips how to access this?

[#2]

Hi @JaNeu ,

  If you really made the DOE with the disallowed combination expression (using JSL, as X1>X2), then the profiler and prediction formula should automatically be undefined for cases where X1>X2. See for example the simulated DOE in the attached data table and associated scripts.

  Or, if you already have a prediction formula and need to add the constraints later, you can go Graph > Profiler and select your prediction formula. Then, you can add the linear restraints later.

Hope this helps,

DS

[#3]

Thanks for the reply.

I see it works perfectly in your example. In my data I have the disallowed combinations script from when i made the design. When I use the Graph > Profiler with a prediction formula from the model, the dissallowed combination script works. Still it does not propagate to the Fit Model functionality. Any chance I can fix it without having to do this loop-around? And would the disallowed combinations alter my prediciton formula if the model had accepted them?

[#4]

Hi @JaNeu ,

  From what I can tell, when running the Fit Model platform, that option is not directly available in the red hot button menu of the Prediction Profiler. However, if you're running the model as described in your DOE, the profiler shouldn't allow you to access regions of the design space where X1>X2. I'm not sure why it appears to be allowing you to do that. In my mock DOE, it doesn't allow you do that. I do not think the disallowed combinations will alter the prediction formula. I believe it basically just makes an "off limit" region in the design space and for any values where X1 is > X2, then it would set those to NaN (not a number). Sorry I can't provide a better answer than that.

Good luck!,

DS