Happy 1999!
I've been celebrating the holidays by (finally) getting to the port of our
S+ NetCDF routines to R, which I'm trying to make into a library. At the
moment, there are two issues I'd appreciate advice on:
1. Is support for mode "single" data planned anytime soon in R? A lot of
our NetCDF files contain "single" data and having to coerce into "double"
will increase the use of R memory drastically (and slow things down).
2. Is there support for "pointer to pointer" arguments to .C? NetCDF files
contain the data attributes, which aren't known until the files are read.
Our S+ code defines a dummy variable for the data dimensions, which are
redefined (calling "S_alloc") and set within the C code once they are known.
The appended example works in S+ using the pointers= argument in .C.
Is there a way to do this in R?
Thanks...Steve Oncley
oncley@ucar.edu
C-code:
void inquire_varid(
int *ncid_p, /* NetCDF file id. File is left open */
int *var_id_p, /* variable id */
long **varlen, /* Length of variable, along each dimension */
long *ndims_r, /* number of dimensions in variable */
/* This argument is created automatically by S+'s
.C for pointers=T variables */
char **vartype) /* character string describing S object type */
{
...
*varlen = dimval = (long *)S_alloc(ndims,sizeof(long));
...
}
S+ call:
...
vardims <- integer(0) # Length of each dimension
vartype <- "" # "integer","single","double" etc
# get info about variable
cdf <- .C("inquire_varid",
cdfid = cdfid,
varid = varid,
vardims = vardims,
vartype = vartype,pointers=c(F,F,T,F))
...
P.S. I'm using R 0.64 under Redhat Linux 5.2.
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
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
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._
.C
3 messages · Steve Oncley, Brian Ripley, Thomas Lumley
On Fri, 1 Jan 1999, Steve Oncley wrote:
Happy 1999! I've been celebrating the holidays by (finally) getting to the port of our S+ NetCDF routines to R, which I'm trying to make into a library. At the moment, there are two issues I'd appreciate advice on:
As people kept telling me, a `package' in R.
1. Is support for mode "single" data planned anytime soon in R? A lot of our NetCDF files contain "single" data and having to coerce into "double" will increase the use of R memory drastically (and slow things down).
Not `soon' from the responses to my requests. I think you exaggerate though: this can at most double the memory usage, and depending on the machine and if you actually compute with the data (rather than move it around) may have negliglible speed penalty. Remember too that anything done with the data inside S-PLUS 3.x (and R, AFAIK) will make several double-precision copies.
2. Is there support for "pointer to pointer" arguments to .C? NetCDF files contain the data attributes, which aren't known until the files are read. Our S+ code defines a dummy variable for the data dimensions, which are redefined (calling "S_alloc") and set within the C code once they are known. The appended example works in S+ using the pointers= argument in .C. Is there a way to do this in R?
Actually, there is no longer a way to do this in S-PLUS: the pointers= argument is not available in S-PLUS 5.x. I think you need to re-think your strategy: an alternative can be seen in the rpart code.
Brian D. Ripley, ripley@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 272860 (secr) Oxford OX1 3TG, UK Fax: +44 1865 272595 -.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.- 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 _._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._
1 day later
On Fri, 1 Jan 1999, Steve Oncley wrote:
Happy 1999! I've been celebrating the holidays by (finally) getting to the port of our S+ NetCDF routines to R, which I'm trying to make into a library. At the moment, there are two issues I'd appreciate advice on: 1. Is support for mode "single" data planned anytime soon in R? A lot of our NetCDF files contain "single" data and having to coerce into "double" will increase the use of R memory drastically (and slow things down).
I don't think there's any immediate plan for genuine single precision support. I have written some code based on an idea of Peter Dalgaard's to simplify porting of code that uses single precision. It uses a vector of integers of class "single" to hold the data and defines various group methods based on coercing back to double precision at the slightest provocation. It's very preliminary at the moment, but you can get it at http://www.biostat.washington.edu/~thomas/single.tgz This obviously wouldn't solve the time or memory problems, but only a few people have time/memory problems that would be solved by single precision support and it appears that none of them has the substantial amount of spare time needed to implement it. Thomas Lumley Assistant Professor, Biostatistics University of Washington, Seattle -.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.- 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 _._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._