[Rd] Issue with aggregate.ts and/or %\% on Windows

[Prof Brian Ripley]

However, in this case it is the use of %/% that is wrong: the fuzz 
ts.eps is supposed to be used.  Will alter (in R-devel for now).

On Tue, 20 Apr 2010, Peter Dalgaard wrote:

> Patrick Aboyoun wrote:
>> I've stumbled across an issue with aggregate.ts that either is due to a
>> misuse of %/% or something deeper relating to numerical precision on
>> Windows. The test code is
>>
>> x <- rep(6:10, 1:5)
>> as.vector(aggregate(as.ts(x), FUN = mean, ndeltat = 5))
>>
>> On Linux and Mac I get the correct answer
>>
>> > x <- rep(6:10, 1:5)
>> > as.vector(aggregate(as.ts(x), FUN = mean, ndeltat = 5)
>> [1]  7.2  8.8 10.0
>>
>> > sessionInfo()
>> R version 2.11.0 RC (2010-04-18 r51771)
>> i386-apple-darwin9.8.0
>>
>> locale:
>> [1] en_US.UTF-8/en_US.UTF-8/C/C/en_US.UTF-8/en_US.UTF-8
>>
>> attached base packages:
>> [1] stats     graphics  grDevices utils     datasets  methods   base
>>
>>
>> and on Windows I get an incorrect answer
>>
>> > x <- rep(6:10, 1:5)
>> > as.vector(aggregate(as.ts(x), FUN = mean, ndeltat = 5))
>> [1] 7.0 8.5 9.5
>>
>> > sessionInfo()
>> R version 2.11.0 beta (2010-04-11 r51685)
>> i386-pc-mingw32
>>
>> locale:
>> [1] LC_COLLATE=English_United States.1252
>> [2] LC_CTYPE=English_United States.1252
>> [3] LC_MONETARY=English_United States.1252
>> [4] LC_NUMERIC=C
>> [5] LC_TIME=English_United States.1252
>>
>> attached base packages:
>> [1] stats     graphics  grDevices utils     datasets  methods   base
>>
>>
>> Walking through the aggregate.ts code I found this difference is due to
>> what 1 %/% 0.2 produces on the different platforms.
>>
>> On Mac and Linux I get
>>
>> > 1 %/% 0.2
>> [1] 5
>>
>> and on Windows I get
>>
>> > 1 %/% 0.2
>> [1] 4
>>
>>
>> I don't know if %/% supports floating point operands so I'm not sure how
>> to report this issue, but here it is anyway.
>
> It's not as straightforward as that. I get 4 on one (32 bit) Linux, and
> 5 on another (64 bit). Based on samples of size one, I wouldn't want to
> conjecture that "bitness" is the root cause, but it is probably not the
> OS per se, rather the CPU or compiler version.
>
> It is even more insidious: I see on the SAME system
>
>> 1%/%0.2
> [1] 4
>> 1/0.2==5
> [1] TRUE
>
> so it isn't just the usual precision issue.
>
> Of course, exact calculations with floating-point numbers is "unsafe at
> any speed",  but this is quite peculiar. Presumably, it comes about
> because of intermediate storage in an extended precision register.
>
>
> -- 
> Peter Dalgaard
> Center for Statistics, Copenhagen Business School
> Phone: (+45)38153501
> Email: pd.mes at cbs.dk  Priv: PDalgd at gmail.com
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595