On Wed, Nov 15, 2006 at 01:22:19PM -0500, YONGWAN CHUN wrote: > I work on large matrices and found something interesting. For multiplication > of matrices, the order has a huge influence on computing time when one of > them is a sparse matrix. In the below example, M is a full matrix and A is a > sparse matrix in a regular matrix class. A %*% M takes much more time than M > %*% A; moreover, t(t(M) %*% t(A)) is much faster than A %*% M with same > result. I would like to know how it is possible. Even though I do not have an > exact reason, this fact may be helpful to others. > > > n <- 1000 > > M <- diag(n) - matrix(1/n,n,n) > > A <- matrix(rnorm(n*n)>2,n,n) > > system.time(M %*% A) > [1] 0.10 0.03 0.12 NA NA > > system.time(A %*% M) > [1] 3.47 0.03 3.50 NA NA > > system.time(t(t(M) %*% t(A))) > [1] 0.23 0.00 0.23 NA NA
Hi Yongwan, Sorry but I could not reproduce this: > n <- 1000 > M <- diag(n) - matrix(1/n,n,n) > A <- matrix(rnorm(n*n)>2,n,n) > system.time(M %*% A) [1] 3.466 0.060 3.744 0.000 0.000 > system.time(A %*% M) [1] 3.327 0.088 4.046 0.000 0.000 > system.time(t(t(M) %*% t(A))) [1] 3.681 0.089 3.814 0.000 0.000 What OS, R version, linalg package are you using? Tamas ______________________________________________ R-help@stat.math.ethz.ch mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
