Understanding and Modeling Response Curves
gail_massari
| Updated on
This video was updated in September 2024.
See how to:
- Explore curve data using Graph Builder
- Fit all basic functions using Functional Data Explorer
- Fit wavelets and examine results
- Fit B- and P- splines
- Examine results of P-spline model using Functional Design of Experiments (FDOE) Profiler
- Use JMP Pro 18 FDE peak finding models to characterize spectra and determine how different inputs impact responses
Questions answered by Clark Ledbetter @Clark_Ledbetter and Jeff Upton @Jeff_Upton at the live session.
Q: I thought Validation rows were held back from model training, and those were a judge of how well the model does on 'unseen data'. What is difference?
A: The Test data is what is withheld from fitting the model. The Validation set is used in conjunction with the Training set to prevent model overfitting. See documentation that explains Test, Train, Validate purposes. for a JMP explanation.
Q: Can this be done using Fit Curve using JMP, if we don’t have JMP Pro or want to use Fit Curve for some reason?
A: Yes, if you have a known mechanistic / physical model use fit curve. If you don't then using JMP Pro FDE would be the capability to use.
Q: Are there ways to load multiple targets? If not, can I export the model so I don’t have to regenerate them each time?
A: it looks like there's only one target function that you're able to specify for those functional responses. At this time you can only handle a single target/term. Regarding saving the model, yes, there is a save Prediction Formulas option right out of the Profiler to save them to the data table. Under the graph profile you probably want to do click the box that says expanded formulas to make sure. If you have X inputs. not actually optimizing response we actually use as a troubleshooting tool.
Q: For example, we may want to generate some bad responses that we know what the known sources are that caused it, then model it. That way if we ever run into similar issues we can just pull that model and identify the likely causes or where to start looking for causes.. How could we handle that?
A: Consider looking at the Olive Oil example described in the second half of this this JMP Discovery presentation. You can watch video and read the transcript to find the olive oil example that might help you. It is part of. I hope it helps.
Questions answered at earlier sessions on this topic:
Q: How would he remove FPC2 if it was determined that it didn’t really add anything to the model?
A: To remove an FPC, the red triangle nex to Functional PCA can be used, along with the option Customize Number of FPC's.
Resources
- Functional Data Explorer documentation
- Peak Finding report documentation
- Mastering JMP video on Identifying the Impact Curved Factor Shapes have on Responses
