[Rd] compairing doubles
Mark van der Loo
m@rk@v@nderloo @ending from gm@il@com
Fri Aug 31 16:00:06 CEST 2018
how about
is_evenly_spaced <- function(x,...) all.equal(diff(sort(x)),...)
(use ellipsis to set tolerance if necessary)
Op vr 31 aug. 2018 om 15:46 schreef Emil Bode <emil.bode using dans.knaw.nl>:
> Agreed that's it's rounding error, and all.equal would be the way to go.
> I wouldn't call it a bug, it's simply part of working with floating point
> numbers, any language has the same issue.
>
> And while we're at it, I think the function can be a lot shorter:
> .is_continous_evenly_spaced <- function(n){
> length(n)>1 && isTRUE(all.equal(n[order(n)], seq(from=min(n), to=max(n),
> length.out = length(n))))
> }
>
> Cheers, Emil
>
> El vie., 31 ago. 2018 a las 15:10, Felix Ernst
> (<felix.gm.ernst using outlook.com>) escribió:
> >
> > Dear all,
> >
> > I a bit unsure, whether this qualifies as a bug, but it is definitly
> a strange behaviour. That why I wanted to discuss it.
> >
> > With the following function, I want to test for evenly space
> numbers, starting from anywhere.
> >
> > .is_continous_evenly_spaced <- function(n){
> > if(length(n) < 2) return(FALSE)
> > n <- n[order(n)]
> > n <- n - min(n)
> > step <- n[2] - n[1]
> > test <- seq(from = min(n), to = max(n), by = step)
> > if(length(n) == length(test) &&
> > all(n == test)){
> > return(TRUE)
> > }
> > return(FALSE)
> > }
> >
> > > .is_continous_evenly_spaced(c(1,2,3,4))
> > [1] TRUE
> > > .is_continous_evenly_spaced(c(1,3,4,5))
> > [1] FALSE
> > > .is_continous_evenly_spaced(c(1,1.1,1.2,1.3))
> > [1] FALSE
> >
> > I expect the result for 1 and 2, but not for 3. Upon Investigation
> it turns out, that n == test is TRUE for every pair, but not for the pair
> of 0.2.
> >
> > The types reported are always double, however n[2] == 0.1 reports
> FALSE as well.
> >
> > The whole problem is solved by switching from all(n == test) to
> all(as.character(n) == as.character(test)). However that is weird, isn’t it?
> >
> > Does this work as intended? Thanks for any help, advise and
> suggestions in advance.
>
> I guess this has something to do with how the sequence is built and
> the inherent error of floating point arithmetic. In fact, if you
> return test minus n, you'll get:
>
> [1] 0.000000e+00 0.000000e+00 2.220446e-16 0.000000e+00
>
> and the error gets bigger when you continue the sequence; e.g., this
> is for c(1, 1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7):
>
> [1] 0.000000e+00 0.000000e+00 2.220446e-16 2.220446e-16 4.440892e-16
> [6] 4.440892e-16 4.440892e-16 0.000000e+00
>
> So, independently of this is considered a bug or not, instead of
>
> length(n) == length(test) && all(n == test)
>
> I would use the following condition:
>
> isTRUE(all.equal(n, test))
>
> Iñaki
>
> >
> > Best regards,
> > Felix
> >
> >
> > [[alternative HTML version deleted]]
> >
> > ______________________________________________
> > R-devel using r-project.org mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-devel
>
>
>
> --
> Iñaki Ucar
>
> ______________________________________________
> R-devel using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>
>
> ______________________________________________
> R-devel using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>
[[alternative HTML version deleted]]
More information about the R-devel
mailing list