[R] order() on vector with Inf's

[ripley@stats.ox.ac.uk]

On Thu, 23 Jan 2003, Murray Jorgensen wrote:

> I have a parameter vector, ibeta, and a corresponding vector of 
> loglikelihoods, llbet. If llbet contains no NAs or Inf's then I can 
> extract the best parameter by
> 
> index <- order(-llbet)[1]

sort.list(llbet, decreasing=TRUE)[1]

would be better.

> beta <- ibeta[index]

or just max(beta, na.rm=TRUE).

> or similar. The argument na.last of order() allows me to fix this up 
> even if llbet contains some NAs which I wish to ignore.
> 
> Unfortunately my llbet contains Inf's, which are not points of infinite 
> likelihood, if anything they should be -Inf's. Anyway na.last does not 
> seem to help me with these. What should I do?

llbet[llbet==Inf] <- -Inf

should correct the problem, but why not correct the calculation?

-- 
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