nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

Functional DoE using Neural Model ?

Hello,

Got a question here:

"FDE" integrate with Function DoE using "Generalized Regression" model which enable us to build a model for the functional factors

Q1: Other than"Generalized Regression", is it possible to build a "Neural Network" or some other models? ( as one of my customers insist to use neural model)

Q2: I'm thinking of using the Function PCA to do a workaround other model builing.

when using "Save Summaries" in the "FDE" platform, the Functional PCA embedded with FDE col. properties but NO formulas,

which seems couldn't link back to the original factors. wondering am I using the correct way or not?

Does anyone have an idea/suggestion on this? many thanks

Accepted answers

Phil_Kay

Just make sure that you have Save Formulas checked in Customize Function Summaries (red triangle menu next to Function Summaries).

Customise Summaries.png

All comments

Mark_Bailey

Yes, you can use the FPCs in the X or Y role in any modeling platform. The inclusion of GenReg was a natural choice because the linear model is so useful in typical DOE analysis. It is a great convenience, but not a requirement. I have not tried a neural network in particular but it should work.

You can supply the explanatory variables in the Z, Supplementary role so that they appear with the FPC when you save the summaries or you an use Table commands to join them after you save the summaries.

Phil_Kay

Just make sure that you have Save Formulas checked in Customize Function Summaries (red triangle menu next to Function Summaries).

Customise Summaries.png

Thank you, Mark. this gives me the idea go-ahead to do neural modeling, it works! just takes a bit more steps to go
Much appreciate

Much appreciate Phil. It's very clear to guide me to next step. with the formulas, I'm able to build a neural model now! Thank you

@Phil_Kay @Mark_Bailey

Use a neural model for FDE sample data “Fermentation Process.jmp” & "Fermentation Process Batch Yield Results.jmp”

Got a profile like this: Yield is a function of FPC of many original factors

went through a few recorded FDE videos, seems all use Generalized Reg. to do the model fitting,

The question is:

1): Is there a way to change the yield (via FPC) back to original factors for the easier interpretation? and also

2): the predictive formula will able to compute using the original factors (rather than using the FPC which need FDE to calculate whenever a new set of data coming) many thanks!

Phil_Kay

Hi,

I think that I understand what you are trying to do. I created a bonus exercise for a workshop that covers this. Look at exercise 2, bonus exercise.

Here are some instructions. These use a model fitted by Gen Reg as the example. But you should be able to do the same with a Neural model, I think.

After you have fit the generalized regression model, click the red triangle next to Normal Lasso with AICc Validation and Save Columns > Save Prediction Formula.
Now Graph > Profiler.
Select the Yield[BatchID] Prediction Formula column and cast into the Y, Prediction Formula role. Also add the “Prediction Formula” columns for Ethanol, Molasses Feed, Air and NH3 Feed. The dialog should look like this:

Profiler FPCs.png

Make sure to click the box for Expand Intermediate Formulas. Click OK.
You might want to rearrange the X-axes: click the red triangle next to Prediction Profiler > Appearance > Reorder X Variables. Drag Time to the top of the list. Click OK.
You might also want to increase the size of the profiler plots by dragging out the corner of one.
Look at Ethanol FPC 2 and how changing the setting affects Yield.
Look at how the Ethanol vs Time profile changes as you change Ethanol FPC 2. Now you can see what ethanol profiles resulting in a low yield (left) and high yield (right) look like…

Ethanol profiles.png