[Rd] compairing doubles

[Serguei Sokol]

Le 31/08/2018 à 16:25, Mark van der Loo a écrit :
> 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
Or with only four function calls:
diff(range(diff(sort(x)))) < eps

Serguei.
>
>
> -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]]
>
> ______________________________________________
> R-devel using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>


-- 
Serguei Sokol
Ingenieur de recherche INRA

Cellule mathématiques
LISBP, INSA/INRA UMR 792, INSA/CNRS UMR 5504
135 Avenue de Rangueil
31077 Toulouse Cedex 04

tel: +33 5 62 25 01 27
email: sokol using insa-toulouse.fr
http://www.lisbp.fr