Setting limits of a factor based on a categorical variable

[#1]

Dear JMP!

We have recently purchased JMP Pro, and I want to utilize Bayesian optimalization.

I would like to generate a starting set of points with a DoE with Fast Flexible Filling Design for the optimalization. 

I have 2 responses and 5 factors, 1 of them is categorical (Dtype), 4 is continuous. 

Based on the level of Dtype, I would like to set the limit of the factor D to different values (e.g. for P and A betweeen 5-10, for L 5-20).

How can I achieve this?

I have attached a screenshot from my JMP Pro.

Thanks in advance for your time and answer.

[#2]

Hi @VarianceCivet85 

Welcome in the Community !

When creating your Space Filling design to generate your candidate set, you can very simply achieve your goal by using Disallowed Combinations Filter or Disallowed Combinations Script.

Select "Use Disallowed Combinations Filter", and select factor Dtype = P and the range for factor D from 10 to 20, OR factor Dtype = A and the range for D from 10 to 20. These are the factors ranges/combination you don't want in your design based on your informations, you should have this screen :

If you are more familiar with JMP script, you can use the option "Use Disallowed Combinations Script" and this script:

Dtype == "P" & D >= 10 | Dtype == "A" & D >= 10

And you'll get the same results.

Specify the number of points you want for your candidate set, generate the datatable, and you can finally check that the constraints are well enforced and respected:

I would recommend reading the excellent blog Demystifying Factor Constraints by @Jed_Campbell, it will explain the different ways to enforce constraints when creating designs/candidate sets.

EDIT: As you want to create starting points for Bayesian Optimization, you could maybe use a D-Optimal Custom Design and enforce the constraint mentioned before, to be sure that the range for each factor can be explored with a minimum of runs. With only 6 runs and Fast Flexible Filling design, I had some situations where the ranges of D for every Dtype was not fully explored : 

With a Custom design approach (and only using main effects in the model), you'll need a minimum of 7 runs, but the repartition of the runs ensure you will explore the entire range for any Dtype level :

Hope this answer will solve your problem,