I've been trying to prepare some wrappers for optimization objective functions and gradients that use try() to check for computable functions. I'm also trying to do scaling as per optim() so that a number of methods that are available on CRAN and R-Forge can use parameter and function scaling. This got me into minor trouble when I scaled a gradient or Hessian and the "inadmissible" attribute I had created suddenly disappeared. When I discovered why -- that some operations don't preserve attributes -- I started to look for a list of which ops preserve attributes and which don't. The code snippet below shows that matrix multiplication (%*%) does not, while multiplication by a vector (simple *) does. I'm not really surprised that some ops don't preserve attributes, particularly those that are binary like multiplication, but it might be nice to know which operations and special functions do so. rm(list=ls()) m<-matrix(1:4,nrow=2, ncol=2) print(m) attributes(m) attr(m,"check")<-"***" attributes(m) bigm<-10*m str(bigm) bigm1<-diag(c(1,1))%*%m str(bigm1) bigm1<-c(1,2)*m str(bigm1) print(bigm1) arraym<-as.array(m) str(arraym) tanm<-tan(m) str(tanm) Best, John Nash
list of operations that preserve attributes
2 messages · John C Nash, Brian Ripley
There is a description in the 'Blue Book' (Becker et al, 1988) of when attributes are supposed to be preserved, and R mostly conforms to it. This is discussed somewhere relevant in the R documentation: I would have to search for exactly where, so I leave that to you.
On Wed, 19 Oct 2011, John C Nash wrote:
I've been trying to prepare some wrappers for optimization objective functions and gradients that use try() to check for computable functions. I'm also trying to do scaling as per optim() so that a number of methods that are available on CRAN and R-Forge can use parameter and function scaling. This got me into minor trouble when I scaled a gradient or Hessian and the "inadmissible" attribute I had created suddenly disappeared. When I discovered why -- that some operations don't preserve attributes -- I started to look for a list of which ops preserve attributes and which don't. The code snippet below shows that matrix multiplication (%*%) does not, while multiplication by a vector (simple *) does. I'm not really surprised that some ops don't preserve attributes, particularly those that are binary like multiplication, but it might be nice to know which operations and special functions do so. rm(list=ls()) m<-matrix(1:4,nrow=2, ncol=2) print(m) attributes(m) attr(m,"check")<-"***" attributes(m) bigm<-10*m str(bigm) bigm1<-diag(c(1,1))%*%m str(bigm1) bigm1<-c(1,2)*m str(bigm1) print(bigm1) arraym<-as.array(m) str(arraym) tanm<-tan(m) str(tanm) Best, John Nash
______________________________________________ R-devel at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-devel
Brian D. Ripley, ripley at stats.ox.ac.uk Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/ University of Oxford, Tel: +44 1865 272861 (self) 1 South Parks Road, +44 1865 272866 (PA) Oxford OX1 3TG, UK Fax: +44 1865 272595