[Rd] compairing doubles

[Mark van der Loo]

Ah, my bad, you're right of course.

sum(abs(diff(diff( sort(x))))) < eps

for some reasonable eps then, would do as a oneliner, or

all(abs(diff(diff(sort(x)))) < eps)

or

max(abs(diff(diff(sort(x))))) < eps


-Mark

Op vr 31 aug. 2018 om 16:14 schreef Iñaki Ucar <iucar using fedoraproject.org>:

> El vie., 31 ago. 2018 a las 16:00, Mark van der Loo
> (<mark.vanderloo using gmail.com>) escribió:
> >
> > how about
> >
> > is_evenly_spaced <- function(x,...) all.equal(diff(sort(x)),...)
>
> This doesn't work, because
>
> 1. all.equal does *not* return FALSE. Use of isTRUE or identical(.,
> TRUE) is required if you want a boolean.
> 2. all.equal compares two objects, not elements in a vector.
>
> Iñaki
>
> >
> > (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
>
>
>
> --
> Iñaki Ucar
>

	[[alternative HTML version deleted]]