275036_en-US

liliconcom

Sequence Simplex optimization or evolutionary optimization

A customer has approached me about a methodology they previously executed via a script to select the best vertices for the next screening/experiment. They now want to employ this technique to find the 'cusp' of a process to optimize good over bad. I've copied their note below. I've looked through the community and haven't been able to come up with a solution. 

 

Is there any add in, script that will enable the use of Sequence Simplex optimization – In one of our plants we have a process in which competing good and bad byproducts are present and will like to use this methodology to find the “cusp” in which we minimize the bad while achieving the highest possible of the good.  I used this methodology in JMP in the past – I remember that started w/ the Custom design in DOE, but had an script that enabled the selection of the best vertices for the next screening/experiment. looking into the community page, I also find that sometimes is called evolutionary optimization.

Thanks for any insight you can offer.

Re: Sequence Simplex optimization or evolutionary optimization

Many years ago when I worked at Dow Chemical, we actually had such a script to perform Simplex Optimization. It did start with a minimum run custom design. Mark Bailey wrote it, and I had to make some modifications to it. I do have the script BUT I know there are problems with it. It never really calculated the reflection properly when there were more than two input variables. I never got around to fixing that part of it. I have attached the original script that Mark and I worked on. Perhaps it can be a good starting point for someone who has the time/inclination to fix the reflection part. I will also start looking at that again, to see if it can be easily fixed.

Re: Sequence Simplex optimization or evolutionary optimization

Also, to clarify, simplex optimization is NOT the same as evolutionary optimization (EVOP). Box wrote an entire book about EVOP. Same goal, very different method.

Re: Sequence Simplex optimization or evolutionary optimization

Hi,

DOW just asked me to update the add-in as the original script uses the legacy overlay platform.

Here is an updated version of the add-in using Graph Builder.

@Dan_Obermiller is this DOW Chemical IP?  

Can we share it?

 

cheers,

Stan

Re: Sequence Simplex optimization or evolutionary optimization

Although Dow Chemical requested the script, they did not pay anything for it. Mark wrote it out of the goodness of his heart! So, I think that means it could be distributed. 

 

So, Stan, did you fix the reflection problem that was in the original script?

Re: Sequence Simplex optimization or evolutionary optimization

Dan,

I was not aware of the reflection problem.

The problem presented to me was "depending on the version of JMP used the plot output was different."

 

Stan

 

Re: Sequence Simplex optimization or evolutionary optimization

Just a little different perspective to what have been said aready:

Taking a look at the topic of evolutionary optimization in combination with DOE, there are papers out there using this optimization technique. Evolutionary Algorithms (EA) are global optimization algorithms allowing to jump out of local extremes, where other stepest decent methods would fail to find the global optima. 

 

That said, they try to find a good range where the optimum probably lays using DOE and then using EA to find the best spot. That is one approach. Another is to use designed experiments to find the best set of points in each iteration of the optimization, instead of random points or certain other mainly factorial/directional approaches for the next generation.

 

It would be important to understand what exactly they are aiming for. If their ultimate goal would be to start with very few treatments, get some results and from there they want to find the next best treatment, you would need to talk about the why as well the costs involved they want to do this kind of sequential approach. Often in such cases there is other information available you will select the next best from. E.g. comparable with material properties used as covariates to select few from many materials to be used in the DOE with the goal to understand the controllable part of the process with best understanding of the dependency on the material.