topic Re: Prediction profiler with mixture factors in Discussions

Prediction profiler with mixture factors

ruskicar — Mon, 11 Dec 2023 22:39:23 GMT

Hi

I am wondering how does Prediction Profiler work when having multiple mixture factors included? When you slide one factor, how are other adjusted?

Is there a way to include some special rules, such as if you decrease certain mixture factor, then due to some cost (or other) limitations you want to force JMP to first try to increase a particular mixture factor, not necessarily equally distribute the gain among all other mixure factors?

Thanks and best regards

Re: Prediction profiler with mixture factors

Jed_Campbell — Mon, 11 Dec 2023 23:14:06 GMT

When you slide one factor, how are other adjusted?

If you slide one factor far enough to violate the mixture constraint, the other factors are forced into a location the satisfies the mixture constraint.

Is there a way to include some special rules, such as if you decrease certain mixture factor, then due to some cost...

Yes. If you include multiple outputs in the model, you can set the relative importance of each output, which will cause the profiler to seek a place that best satisfies what you're asking for when you select "Maximize Desirability." In the example below, cost is only half as important as the output.

Re: Prediction profiler with mixture factors

Victor_G — Tue, 12 Dec 2023 08:26:29 GMT

Hi @ruskicar,

@Jed_Campbell did answer all your points. I just want to add a complementary comment on your question :

Is there a way to include some special rules, such as if you decrease certain mixture factor, then due to some cost (or other) limitations you want to force JMP to first try to increase a particular mixture factor, not necessarily equally distribute the gain among all other mixure factors?

On the Profiler, you can fix some factors at specific values if you don't want the profiler to move the value during optimization or just in the interaction. This is particularly useful for mixture designs when you want to assess how some of the mixture factors affect the responses, by fixing one of the component to a certain value.

In order to do this, use CTRL + left click on the mixture factor (in the profiler panel) you want to fix, specify the value in "Current Value" and check the option "Lock Factor Setting" :

This way you can play with the Profiler but this factor will stay at its fixed value.

You can also use the "Mixture Profiler" (red triangle, "Factor Profiling", "Mixture Profiler") to move in the experimental space more easily and flexibly than with the Prediction Profiler :

Hope this response will help you,