Yes, Bryan put irlba up on CRAN. It works on either matrix or big.matrix objects, too. However (as Bryan would note) it is a pure R implementation and its performance could be greatly improved by moving much of the algorithm to C (maybe I'll try to get a grad student interested in working on it). So it might not be surprising if the performance in "small-medium" simulations is unimpressive. Just remember: it can do a *really big* partial SVD, which most other alternatives (that I'm aware of) wouldn't be able to handle. We've used it on an 80 GB data set (Netflix data without using a sparse representation) successfully. Jay
John W. Emerson (Jay) Associate Professor of Statistics Department of Statistics Yale University http://www.stat.yale.edu/~jay