Choose Language Hide Translation Bar
Highlighted
Victor_G
Level III

Possible linking between FDE profiler and mixture profiler

Hello dear JMP-users,

I have a question regarding the use of FDE profiler and mixture profiler.

I have two DoE datatables, one with summarized tests results from a mixture design with 3 components (giving me only one value for the response), and the other one with complete tests results from the same DoE (with signal curves as test response).

 

Problem : When I use the DoE-FDE profiler to look at the changes in the curve, I struggle to look at the different possibilities of mixtures, as changing one component quantity will automatically change other components quantities. This is normal because of the mixture design (and dependancy of the mixture factors), but I have troubles to have a curve for a mixture at X1 = 0,6 ; X2 = 0,3 and X3 = 0,1 for example, because when I change X1 to 0,6 JMP change automatically quantities of X2 and X3 at 0,2.

 

Is it possible to link the mixture profiler from the summarized datatable (triangle) with the DoE-FDE profiler ? In order to point specific mixtures in the triangle/mixture profiler and look at the corresponding curves in the FDE profiler ?

Thanks in advance for your help ! 

Victor G.

2 ACCEPTED SOLUTIONS

Accepted Solutions
Highlighted

Re: Possible linking between FDE profiler and mixture profiler

That's good - the latest version includes a lot of enhancements to make what you want to do very easy!

 

Unfortunately, the Mixture Profiler does not work with functional data models. You can still use other profilers, though, such as the Prediction Profiler. In case that approach is useful but you are not familiar with the steps involved, follow these steps:

 

You must use the stacked column format to be able to put your design and the functions together. So one column for the functions, one column for the domain, and one column for the function identity. Add a column to the design data table that associates the experiment run (row) with the function ID. Then use Tables > Join to merge them by matching on the ID column in each data table.

 

Now select Analyze > Specialized Modeling > Functional Data Explorer. Use the default tab in the launch dialog box for stacked data. Assign the functions, domain, and ID as usual. Select the factor data columns and assign them to the Z, Supplementary role. The last step is key to making this analysis easy.

 

Fit your basis expansion model and select the number of FPCs that you want as before. Click the red triangle next to Function Summaries and select Save Summaries. This command produces a new data table. It automatically includes a table script to launch the Prediction Profiler.

 

I hope that this answer is helpful, even if you can't get the profiler that you want.

Learn it once, use it forever!

View solution in original post

Highlighted
Victor_G
Level III

Re: Possible linking between FDE profiler and mixture profiler

In this case the aim is exploration, and figure out what certain quantities change in a mixture can affect the signal (and where).
I think I figure out the solution :
By opening the two tables, with the ID column identical for both of them (and same column names for mixture factors), I opened the "normal model report" on one side (with the "summarized" datatable to have the mixture profiler) and the Functional Data Explorer on the other side (with the "complete" datatable). I click in the Mixture profiler on "Link Profilers".
When I'm changing the point in the experimental space in the mixture profiler, it automatically changes the factor concentrations and curves in the functional report !

View solution in original post

11 REPLIES 11
Highlighted

Re: Possible linking between FDE profiler and mixture profiler

Are you using JMP Pro version 14 or 15?

Learn it once, use it forever!
Highlighted
Victor_G
Level III

Re: Possible linking between FDE profiler and mixture profiler

Hello @markbailey !
I'm using JMP Pro 15. 

Highlighted

Re: Possible linking between FDE profiler and mixture profiler

That's good - the latest version includes a lot of enhancements to make what you want to do very easy!

 

Unfortunately, the Mixture Profiler does not work with functional data models. You can still use other profilers, though, such as the Prediction Profiler. In case that approach is useful but you are not familiar with the steps involved, follow these steps:

 

You must use the stacked column format to be able to put your design and the functions together. So one column for the functions, one column for the domain, and one column for the function identity. Add a column to the design data table that associates the experiment run (row) with the function ID. Then use Tables > Join to merge them by matching on the ID column in each data table.

 

Now select Analyze > Specialized Modeling > Functional Data Explorer. Use the default tab in the launch dialog box for stacked data. Assign the functions, domain, and ID as usual. Select the factor data columns and assign them to the Z, Supplementary role. The last step is key to making this analysis easy.

 

Fit your basis expansion model and select the number of FPCs that you want as before. Click the red triangle next to Function Summaries and select Save Summaries. This command produces a new data table. It automatically includes a table script to launch the Prediction Profiler.

 

I hope that this answer is helpful, even if you can't get the profiler that you want.

Learn it once, use it forever!

View solution in original post

Highlighted
Victor_G
Level III

Re: Possible linking between FDE profiler and mixture profiler

Hello @markbailey ,
Thanks a lot for this interesting outcome, which enables to have access to curves easily for existing test points in the mixture design.
I wish it would be possible to link one mixture profiler with this functional profiler, in order to have the predicted curves for any point in the mixture design.
I will try something else with the ID column and the "ID Link" or "ID reference"  column properties.

Thanks a lot for the time spent ! 
 

Highlighted

Re: Possible linking between FDE profiler and mixture profiler

If you think about it, it doesn't make sense to want to put functions and the Mixture Profiler together. This profiler is for the response surfaces. That is, a single response value for any combination of component proportions. The Prediction Profiler, though, has been nicely adapted to profile the curve (a profile, right?) as a function of the components. It could also include process factors with mixture components.

 

In case it is not clear, the Functional DOE feature in the FDE fits a model of the functions to the mixture components through the FPCs to give you a Prediction Profiler like this:

 

Capture.JPG

Learn it once, use it forever!
Highlighted
Victor_G
Level III

Re: Possible linking between FDE profiler and mixture profiler

I understand the problem behind the two data types.
My idea behind would be to have a mixture profiler with a single value (mean value from a certain area in the signal) as a response to create the mixture profiler, and use this profiler (link it to another datatable for example, with functional data but the same mixture factors X's) only to change the mixture X's factors for the Functional prediction profiler (I'm not interested in the surface response for the mean value response, only in the visualization of the experimental domain for the FDE). Because the actual version that I get is not "user-friendly" for the changes in the mixture X's factors, as I'm not able to specify easily a particular spot in the triangle where I would like to see a predicted curve.
But if it is not possible, that's ok, I just didn't wanted to miss this option
Thanks for the support !
Highlighted

Re: Possible linking between FDE profiler and mixture profiler

So it seems that you want to optimize a scalar response, not a functional response. That is, the curve shape is not important, just the response at a particular domain value.

Learn it once, use it forever!
Highlighted
Victor_G
Level III

Re: Possible linking between FDE profiler and mixture profiler

In this case the aim is exploration, and figure out what certain quantities change in a mixture can affect the signal (and where).
I think I figure out the solution :
By opening the two tables, with the ID column identical for both of them (and same column names for mixture factors), I opened the "normal model report" on one side (with the "summarized" datatable to have the mixture profiler) and the Functional Data Explorer on the other side (with the "complete" datatable). I click in the Mixture profiler on "Link Profilers".
When I'm changing the point in the experimental space in the mixture profiler, it automatically changes the factor concentrations and curves in the functional report !

View solution in original post

Highlighted

Re: Possible linking between FDE profiler and mixture profiler

You can lock the value of a factor in the Prediction Profiler, Control-click with Windows or command-click with Mac on the profile for a factor and click the check box for Lock. You can change it to any value you want but JMP can't any longer. Now change another one. And so on.

Learn it once, use it forever!
Article Labels

    There are no labels assigned to this post.