On Tue, 2005-05-31 at 11:11 -0400, Duncan Murdoch wrote:
> Marc Schwartz wrote:
>
> > Final note to Henrik: Note that the IEEE 754 rounding standard as
> > implemented in R results in:
> >
> >
> >>round(18.15, 1)
> >
> > [1] 18.1
> >
> >>formatC(18.15, format = "f", digits = 1)
> >
> > [1] "18.1"
> >
> >>sprintf("%5.1f", 18.15)
> >
> > [1] " 18.1"
> >
> > This is because the rounding method implemented is the "go to the even
> > digit" approach. Thus, you don't get 18.2.
> >
> > See ?round for more information.
>
> I don't think "go to the even digit" is being applied here: ".1" is not
> an even digit.
>
> I suspect what's going on in this example is that 18.15 is not being
> represented exactly; it's stored internally as something slightly less
> than that value, so it rounds down.
>
> You'd see the "go to the even digit" rule applied when rounding 17.5 or
> 18.5, which can be represented exactly, being fractions with a power of
> 2 in the denominator:
>
> > round(18.5, 0)
> [1] 18
> > round(17.5, 0)
> [1] 18
>
> (This is very gratifying. Usually when I try to predict the exact
> behaviour of round() or signif() I end up having to rewrite my
> prediction afterwards. But this time I got it right. Honest!)
>
> Duncan Murdoch
Duncan,
Just got back from a day long meeting.
You are indeed correct on the rounding here. If you look at how 18.15
appears when printed with more significant digits:
> print(18.15, 20)
[1] 18.149999999999998579
That's what I get for trying to deal with floating point representation
issues first thing after a three day weekend... ;-)
Thanks for the correction.
Marc
______________________________________________
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
