What inspired this Wish List Request? Functional data analysis is great, but when using functional Principal Components, you need to save out the functional principal components to do a functional PCR to build a predictive model because it is an unsupervised method. In theory functional Partial Least Squares would remedy this because it is a supervised method and building predictive models would be much easier.
The original inspiration came from a poster that Peter Hersh and I presented at European Discovery and more recently at the SciX conference. We compared PCA, functional PCA, and PLS for analyzing spectral data. PLS came out on top with functional PCA coming in second. So, if functional PCA is better than PCA, it would stand to reason that functional PLS would be better than PLS, at least in my mind. JMP would then have the best, most predictive modeling capabilities built right into the FDE platform. No other software has this capability unless you want to code it in R or Python.
Please describe the new or improved feature or solution you would like to see. Adding functional PLS to the FDE platform would make JMP's FDE a force to be reckoned with for analyzing curve data, and it would be a more flexible platform for analyzing any curve data. Please see the attached paper from Taylor and Francis on behalf of the American Statistical Association that directly compares Functional Principal Component Regression and Functional Partial Least Squares. The paper describes and compares functional PCR and functional PLS and functional PLS comes out on top for analyzing real world NIR data.
What would the value be to you and/or other users if this new feature is provided? Functional PLS would give a more robust and direct way of analyzing curve data. Having functionalPLS would make it much easier to build predictive models within the FDE platform which is something users have been asking for since FDE was introduced.
Is this an external customer request? Not in the sense that a customer has asked for, but customers have been asking for a way to directly build predictive models from functional PCA data. I did speak with Drew Luebe of ADM about adding functional PLS to FDE, and he is very interested in being able to apply it to their curve data.
Is there a competitive software package that has this feature or solution that we should be aware of? No, and that is the best part. That would make the FDE platform that much more unique in the world of analyzing curve data.
Will this impact a renewal, expansion, or adoption at a customer, and does this put an account at risk if we do not address this? Adding functional PLS to the FDE platform will drive the expansion and use of JMP Pro within our core chemical industry customers because they will be better equipped for analyzing chemometric data.
