Multi-way PCA

Question

A customer recently asked if JMP can do a three-way PCA like what is discussed in the attached article.  The authors used a Tucker3 algorithm in Matlab to do PCA on a multi-way data arrays.  I didn't find anything in JMP documentation so I think the answer is, no, JMP can't do this.  I am hoping to confirm this conclusion and see if anyone has any insight about this type of analysis.

Thanks! -Andrea

3-Way PCA.pdf

andreacoombs · Answer

It sounds like the Multiple Factor Analysis platform in Consumer Research is the, or an, answer.  Although it doesn't use the Tucker3 algorithm, the idea is similar.  It can can fit PCA models to the groups of variables separately then combine them into a global PCA.  The Wine Sensory Data in the sample data is a good example (Link to MFA wine example in JMP Documentation).  The Consensus Map will show a scatter plot of the loadings to know which wines (or noodle formulations as in the paper) are similar and the Biplot will show you if you have differences between judges.  You can even include supplementary variables and see if the wines (based on sensory data) cluster to together based on the supplementary variables.  In the example, the supplementary variables represent region but I can see these being data from their roasting process - even fPCs from FDE!!  This could be really interesting conversation and Discovery Paper @Peter_Hersh - and a good reason to be jealous @Jason_Wiggins ;) .

Peter_Hersh · Answer

@andreacoombs That sounds like a great Discovery paper.  I hope it works out.

Jason_Wiggins · Answer

Feeling some jealousy Andrea! Sounds like a really fun problem. Let us know what they end up doing:)