Sparse Linear Algebra for the R Language
Pin Ng, (Northern Arizona University), Pin.Ng@nau.edu, and
Roger Koenker, (University of Illinois), firstname.lastname@example.org
Many contemporary applications in statistics involve large sparse matrices, matrices with a high proportion of zero entries. Conventional array storage and associated basic linear algebra routines can be extremely burdensome for such matrices. However compressed storage schemes and specially designed algorithms can drammatically improve performance. In this talk, we will describe an implemenation of sparse linear algebra methods for the statistical language R. The implementation relies heavily on Saad's (1994) Sparskit package and the Cholesky factorization algorithms of Ng and Peyton (1993). Some performance aspects will be illustrated with applications to sparse linear regression problems including penalized L1 and L2 regression for smoothing problems. Sparse matrix classes are also useful in transferring data across platforms and software environments. A variety of sparse storage formats will be described and conversion methods discussed.