>>>>> Martin Maechler on Thu, 2 Jan 2025 20:42:58 +0100 writes:
>>>>> Duncan Murdoch on Thu, 2 Jan 2025 11:28:45 -0500 writes:
>> On 2025-01-02 11:20 a.m., Duncan Murdoch wrote:
>>> On 2025-01-02 9:04 a.m., Norbert Kuder wrote:
>>>> Hello all,
>>>>
>>>> I am running R version 4.4.2 (2024-10-31 ucrt) on Windows 10 x64, and
>>>> noticed something that might be a minor bug (or at least inconsistent
code)
>>>> in the stats/arima.R package.
....................
>>>> 2. An unused 'mod' variable assignment at line 190:
>>>>
>>>> mod <- makeARIMA(trarma[[1]], trarma[[2]], Delta, kappa, SSinit)
>>>>
>>>> I am trying to confirm whether this is intended behavior or possibly an
>>>> overlooked detail. Could someone please clarify if the current logic is
>>>> correct?
>> Mystery solved: code like this appears several times in that file. In
>> the occurrence here:
>>
https://github.com/wch/r-source/blob/4a1ed749271c52e60a85e794e6f34b0831efb1ae/src/library/stats/R/arima.R#L293
>> it does appear that the mod value isn't being used.
>> Duncan Murdoch
> Thank you, Norbert and Duncan.
> A little bit (unfinished) aRcheology showed that both parts
> have been in the arima code since Dec 11 2003 (when the code, i.e., the
> whole package 'ts') was moved / merged into package 'stats'.
> I'll fix and quickly test the change, and then commit it.
and testing quickly revealed what is obvious in hindsight:
When minimizing the negative log likelihood, in
optim(<p>, armafn, ...)
the armafn() is called of course and it *does* need the model,
i.e., `mod`.
==> the 2nd "inconsistency was *not* a mistake at all: The
mod <- makeARIMA(....)
line has clearly been necessary all along.
Martin
______________________________________________
[email protected] mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
