[R] julian() and numerical noise

[Andreas Eckner]

Very helpful, thanks!

Andreas

On 3/24/2012 10:57 AM, R. Michael Weylandt wrote:
> It might be easier to just work with a character representation rather
> than fighting the floating point demons:
>
> fnc<- function(date){
>      as.Date(paste(date, "01", sep = "."), format = "%Y.%m.%d")
> }
>
> Hope that helps,
>
> Michael
>
> On Sat, Mar 24, 2012 at 10:47 AM, Andreas Eckner<andreas at eckner.com>  wrote:
>> Thanks for the help. A workaround seems indeed easy to write, for example,
>> by rounding the the value of 'm' inside chron:::julian.default(), or only
>> passing integers to the function.
>>
>> I never intended to use a case involving a quadrillionth of a day, but
>> encountered the problem when converting a dates in YYYY.MM format to a CHRON
>> object. Here is an example for future reference (the output might be highly
>> system-dependent):
>>
>>>    library(chron)
>>>    date<- 2012.02
>>>    y<- floor(date)
>>>    m<- (date %% 1) * 100
>>>    julian(2, 1, 2012) - julian(m, 1, y)
>> [1] 1
>>
>> ~Andreas
>>
>> P.S.: I have included Kurt Hornik on the email.
>>
>> On 3/23/2012 10:52 PM, R. Michael Weylandt wrote:
>>> julian() is in the package base. chron only provides julian.default to
>>> avoid the error Uwe observed. If you look at the code for
>>> julian.default, it's not too hard to see why date will be slightly
>>> more sensitive to "numeric fuzz"....whether that's intentional (or
>>> even a good/bad thing) is somewhat hard to say, but I'd guess the
>>> maintainer didn't really consider a use case involving one
>>> quadrillionth of a day.
>>>
>>> Is it problematic for you? If so, a workaround is not hard to provide
>>>
>>> Michael
>>>
>>> On Fri, Mar 23, 2012 at 4:17 PM, Andreas Eckner<andreas at eckner.com>
>>>   wrote:
>>>> Sorry, forgot to mention: R version 2.14.2
>>>>
>>>> The function julian() is part of the "chron" package of the base
>>>> distribution.
>>>>
>>>> On 3/23/2012 2:47 PM, Uwe Ligges wrote:
>>>>> On 23.03.2012 15:45, Andreas Eckner wrote:
>>>>>> Hi,
>>>>>>
>>>>>> does anybody know if the following behavior of julian() is intentional?
>>>>>>
>>>>>>> julian(2, 1, 2012) - julian(2 - 1e-15, 1, 2012)
>>>>>> [1] 1
>>>>>>> julian(2, 1, 2012) - julian(2, 1 - 1e-15, 2012)
>>>>>> [1] 0
>>>>>>> julian(2, 1, 2012) - julian(2, 1, 2012 - 1e-15)
>>>>>> [1] 0
>>>>>>
>>>>>> In other words, julian() is subject to numerical noise in the 'day'
>>>>>> argument, but not in the 'month' and 'year' argument? Another example:
>>>>>>
>>>>>>> julian(2 - 1e-15, 1, 2012) - julian(1 - 1e-15, 1, 2012)
>>>>>> [1] 30
>>>>>
>>>>>
>>>>> Which vbersion of R are you talking about?
>>>>>
>>>>> I get:
>>>>>
>>>>>> julian(2, 1, 2012) - julian(2 - 1e-15, 1, 2012)
>>>>> Error in UseMethod("julian") :
>>>>>   no applicable method for 'julian' applied to an object of class
>>>>> "c('double', 'numeric')"
>>>>>
>>>>>
>>>>> Uwe Ligges
>>>>>
>>>>>
>>>>>> Thanks,
>>>>>> Andreas
>>>>>>
>>>>>> ______________________________________________
>>>>>> R-help at r-project.org mailing list
>>>>>> https://stat.ethz.ch/mailman/listinfo/r-help
>>>>>> PLEASE do read the posting guide
>>>>>> http://www.R-project.org/posting-guide.html
>>>>>> and provide commented, minimal, self-contained, reproducible code.
>>>>>
>>>>>
>>>> ______________________________________________
>>>> R-help at r-project.org mailing list
>>>> https://stat.ethz.ch/mailman/listinfo/r-help
>>>> PLEASE do read the posting guide
>>>> http://www.R-project.org/posting-guide.html
>>>> and provide commented, minimal, self-contained, reproducible code.
>>>
>