Distribution and quantile precision (was [R] Underflow warnings?)

John Logsdon j.logsdon at lancaster.ac.uk
Mon Feb 14 16:49:03 CET 2000

On Sun, 13 Feb 2000, Thomas Lumley wrote:

> A bigger issue, which we _have_ been working on, is getting closer to full 
> precision for the probability and quantile functions. 

Ah - will this include functions for the logs of densities and quantiles?
I recently fell flat on my nose when, in the middle of a very long (in
time) calculation I was suddenly calculating log(dnorm) where the variance
was reducing (as it should) without thinking too much ('cos it was easier
to swap distributions).  This meant that the (x-xbar)/sigma was increasing
to the point where it all blew up.  Mea culpa I know.

I recoded it appropriately and it worked fine.  Kept the chips burning all
Christmas though.

So for those of use that regularly use maximum likelihood, it would be
nice to have log(pdist), log(1-pdist) and log(ddist) where dist is as you
like it - norm to start with.


r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch

More information about the R-help mailing list