Skip to content

p./d./q./r: S compatibility and time-series software for R?

3 messages · Martin Maechler, Paul Gilbert, Kurt Hornik

Original
Martin Maechler
# 
Martyn> I will do this if you haven't been overrun with volunteers
    Martyn> already ;) 
No, you are the first to answer.
Thank you very much in advance!

---------------------

    Martyn> Some time ago you asked about a basic time series
    Martyn> library for R.  Did anything come of that?

Well, yes and no (more "no"):

1. Paul Gilbert has made his DSE (Dynamic Systems Estimation) library
   available for testing on  <<CRAN>>/src/contrib/devel/
   Unfortunately, I haven't yet had time to really look at it.

2. Ross Ihaka:
   Used to have a master's student looking at time series software.
   During the power failure break in Auckland (about 2 weeks ago),
   Ross said he was going home (because there was no electricity at his
   office) and would do some work on time-series stuff.

   He used to have time-series things (for R) in his private 
   directories,  and some of us have been begging him, to also
   ``release'' this could into <<CRAN>>/src/contrib/devel/ .
   Unfortunately, this has not yet happened.

3. Kurt Hornik has ported David James'  "chron" library
   (for chronological objects)
   and als put into <<CRAN>>/src/contrib/devel/  (why "devel"?).
   [Currently, the URL given in the above README doesn't work anymore..?]

4. Several other people seem to use some time-series R functions of their
   own.  There's no coordination and no common format / class, etc.

---
I think the important point here is to adopt a general common framework for
doing time-series in R.
The "tframe" (?) class in 'DSE' may be the most general and the way to go.
On the other hand, it looks somewhat complicated and delicate to work with.

Anyway, I think it would be a pain if we had several non-compatible classes
for time-series objects..
Also, S-plus compatibility (for some functions, not for all) may be an
issue.

Do we have a volunteer here?
-------
Martin
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._
Paul Gilbert
# 
As far as I know my time series package (dse) is working fairly well in R. There
are a few rough spots in the install and I haven't documented these very well,
so I haven't moved it out of devel, but it does seem to be working ok. There are
two things I use in Splus which I have only partly worked around in R.

1/ ar - I have a test version of this but it does not reproduce Splus results

2/ acf - I have type="correlation" working, but not type="partial"

I won't get to these for a long time, so if anyone is interested in working on
them I would be happy to provide what I have.

As far as other time series and related things are concerned:
-my package does not have much in the way of frequency domain, though I have
played with it a little bit and there are a few undocumented things that
possibly don't even work properly. If anyone is interested in working on this
then please contact me and I will provide what I have.
-I would really like to have the ability to use the same random number generator
in S and R, as lots of my work involves simulations and I like to be able to run
a common set of tests to make sure things aren't messing up. Martin has provided
me with code but I haven't had time to look at it. There do seem to be some
issues related to how one should specify what generator to use (e.g. an
environment variable is probably the easiest, but then reproducibility depends
on saving environment settings).

And beyond Splus compatibility:
-I have a set of programs for doing multivariate optimization (nonlinear simplex
and DFP) which I (would like to) use for mle. My use is for time series models,
but I  have it organized as an independent package. This is code which I wrote a
fair time ago, but put aside until the memory problems in S were fixed. I am
fairly sure this will work reasonably in R, but there are still a few bugs in
the code. If anyone is interested in working on this I can provide more details
and some assistance.
-I am starting to work on putting neural net time series models into my package.
I am playing with two possibilities: an interface to the SNNS package, and doing
it all in R. (This does actually seems to be possible in R, whereas in S it is a
disaster.) This is a new area for me, so things will be going slowly. If anyone
is interested in collaborating please let me know.

Paul Gilbert


-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._
Kurt Hornik
# 
Martyn> I will do this if you haven't been overrun with volunteers
Martyn> already ;)
Martyn> Some time ago you asked about a basic time series
Martyn> library for R.  Did anything come of that?
Yes, Ross, please :-)
Because it uses Inf and NaN, which are only available under 0.62.
Yes, it's

http://netlib.bell-labs.com/cm/ms/departments/sia/project/misc/chron.html

now.  I'll make the change.

-k
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._