R devel - Dec 2018 - Bug report: Function ppois(0:20, lambda=0.9) does not generate a non-decreasing result.

function ppois is a function calculate the CDF of Poisson distribution, it
should generate a non-decreasing result, but what I got is:

> any(diff(ppois(0:19,lambda=0.9))<0)
[1] TRUE


Actually,


> ppois(19,lambda=0.9)<ppois(18,lambda=0.9)
[1] TRUE



Which could not be TRUE.


Code is tested in both R 3.5.1 and Microsoft R Open 3.5.1.


               _                          
platform       x86_64-w64-mingw32         
arch           x86_64                     
os             mingw32                    
system         x86_64, mingw32            
status                                    
major          3                          
minor          5.1                        
year           2018                       
month          07                         
day            02                         
svn rev        74947                      
language       R                          
version.string R version 3.5.1 (2018-07-02)
nickname       Feather Spray      







PS. if set lambda=1, such bug will not appear.








	[[alternative HTML version deleted]]

On Tue, 4 Dec 2018 at 11:12, <qweytr1 at mail.ustc.edu.cn>
wrote:
>
> function ppois is a function calculate the CDF of Poisson distribution, it
should generate a non-decreasing result, but what I got is:
>
> > any(diff(ppois(0:19,lambda=0.9))<0)
> [1] TRUE
>
> Actually,
>
> > ppois(19,lambda=0.9)<ppois(18,lambda=0.9)
> [1] TRUE
>
> Which could not be TRUE.
This is just another manifestation of

0.1 * 3 > 0.3
#> [1] TRUE

This discussion returns to this list from time to time. TLDR; this is
not an R issue, but an unavoidable floating point issue. Solution:
work with log-probabilities instead.

any(diff(ppois(0:40, lambda=0.9, log.p=TRUE))<0)
#> [1] FALSE

I?aki

Le 04/12/2018 ? 11:27, I?aki Ucar a ?crit?:
> On Tue, 4 Dec 2018 at 11:12, <qweytr1 at mail.ustc.edu.cn> wrote:
>> function ppois is a function calculate the CDF of Poisson distribution,
it should generate a non-decreasing result, but what I got is:
>>
>>> any(diff(ppois(0:19,lambda=0.9))<0)
>> [1] TRUE
>>
>> Actually,
>>
>>> ppois(19,lambda=0.9)<ppois(18,lambda=0.9)
>> [1] TRUE
>>
>> Which could not be TRUE.
> This is just another manifestation of
>
> 0.1 * 3 > 0.3
> #> [1] TRUE
>
> This discussion returns to this list from time to time. TLDR; this is
> not an R issue, but an unavoidable floating point issue.
Well, here the request may be interpreted not as "do it without round 
error" which is indeed unavoidable but rather "please cope with
rounding
errors in a way that return consistent result for ppois()". You have 
indicated one way to do so (I have just added exp() in the row):

any(diff(exp(ppois(0:19, lambda=0.9, log.p=TRUE))) < 0)
#[1] FALSE

But may be there is another, more economic way?

Serguei.
>   Solution:
> work with log-probabilities instead.
>
> any(diff(ppois(0:40, lambda=0.9, log.p=TRUE))<0)
> #> [1] FALSE
>
> I?aki
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>

-- 
Serguei Sokol
Ingenieur de recherche INRA

Cellule math?matiques
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 at insa-toulouse.fr
http://www.lisbp.fr