skh@nl.cs.cmu.edu (Steve Handerson) (09/20/90)
Folks, I've discovered what I really need is something referred to/invented by Harshman, R.A. and Lundy, M.E. in PARAFAC, or another algorithm that finds an approximate decomposition using only the K largest eigenvalues. Sorry if people are getting tired of this, but I'll justify it by saying anybody doing information retrieval might want these/this reference(s). Here are the two (very similar) references: "Data preprocessing and the extended PARAFAC model" in H.G.Law, C.W.Snyder, Jr., J.A.Hattie, and R.P.McDonald (Eds) "Research Methods for Multimode Data Analysis", Praeger, 1984b. "The PARAFAC model for three-way factor analysis and multi-dimensional scaling" in same, 1984a. Thanks everyone for the response to my SVD post. I'm sure I have at least two working copies of code to choose from. [Most people pointed out Numerical Recipes, but I also got a converted LINPACK routine and a reference to a supposedly faster R-SVD algorithm due to Chan] It's just that the algorithms tend to be K cubed... -- Steve