[GRASS-stats] Using Rcpp to interface GRASS from R and vice versa

[Roger Bivand]

On Wed, 18 Jul 2012, Vaclav Petras wrote:

> Hello grass-stats,
>
> I've tried the R packages Rcpp [1] and inline [2] to write a function
> which returns values from a GRASS raster map using GRASS C functions.
> In attached demo function returns only one value but more complex
> objects can be returned too. There is also an option to pass C++
> classes and R data frames.
>
> In general Rcpp (+inline) can be used to write efficient C++ code and
> use it in R. In case of GRASS and R it can be used to access large
> raster datasets from R without the need to load everything to memory.
>
> There is also another possibility. One can use Rcpp with RInside
> package [3] and write C/C++ code which calls R internally. So GRASS
> module which does some specific computations in R but handles memory
> itself can be created. Note that RInside is R instance within C++
> program, it can run any R code.
>
> I'm sending a very simplified example how things are working. For
> running it, you need latest R, Rcpp package and inline package [4].
> Tested only on GNU/Linux (Ubuntu) but Rcpp should work even on MS
> Windows.
>
> I'm interested (as an programmer) in this topic and I would like to
> write something which uses these possibilities. I'm not experienced
> with R so I'm not sure which way to go. So there are several
> questions.
>
> * Would it be useful to have something like R raster package [5] but
> with GRASS raster maps as data format (direct access to GRASS maps
> from R within GRASS session)?
>
> * Should some GRASS modules or even wxGUI provide some R functionality?
>
> * Not sure, but I think that something like R-based raster calculator
> can be also created. Would be this helpful?
>
> * Why are Rcpp and inline not used in spgrass6 [6]? Is there some
> other reason than Rcpp is relatively new so it may be considered as
> unstable?

The original GRASS package for R shipped with a frozen copy of the GRASS 
5.* libraries as source code. This was not a good idea, and was very hard 
to maintain. Linking against GRASS 6.* and 7.* libraries has not been seen 
as being easy to maintain either, hence the use of GDAL as an interface 
via intermediate files. As you probably know, Rcpp has only been around 
for a short time, and additionally, GRASS is not written in C++, so that 
many of the benefits of Rcpp in handling classes, including S4 reference 
classes in R, are not exploited. spgrass6 needs to run cross-platform, and 
has managed this even where GRASS is 64-bit and R 32, or vice-versa. 
Linking would make this more difficult for users.

Writing a loosely-coupled interface between the raster package via rgdal 
and spgrass6 (or using r.*.bin on the GRASS side) would be of potential 
value, but so far users haven't been asking for it. I don't see that Rcpp 
would help in general terms here, but it might be interesting to see use 
cases and alternative implementations.

I can see that putting some R functionality inside GRASS using RInside may 
be interesting, such as the v.krige addon, getting to R via Python, but 
because the GRASS GUI also matters, it may be that going through Python is 
as good a choice. Again, use cases and alternative implementations would 
be needed to see which routes are helpful.

GRASS functionality is available in R through execGRASS() in spgrass6, 
which apes the interface descriptions exposed by GRASS modules for the 
GUI; it only uses the R system(), so is loosely coupled, but has turned 
out to be quite flexible and powerful.

This may seem depressing, but stability, cross-platform implementation, 
and maintainability really do matter. It isn't impossible that Rcpp could 
help, as could RInside, but it is much more attractive in setting where 
the code to be linked to R is a C++ library, probably a smallish one, with 
no other interaction with the user. GRASS isn't C++, is a collection of 
modules, and does interact with the user in other ways, like the GUI(s).

Thanks for a stimulating post!

Roger


>
> * Would it be more useful to bring some functionality from GRASS to R
> or in the opposite way, trying to use Rcpp within spgrass6 (or
> spgrass7) or just to have some general tools or just to create a guide
> how to combine things?
>
>
> Comments and suggestions are welcome.
>
> Vaclav Petras
>
> [1] http://cran.r-project.org/web/packages/Rcpp/
> [2] http://cran.r-project.org/web/packages/inline/
> [3] http://cran.r-project.org/web/packages/RInside/
> [4] http://cran.r-project.org/bin/
> [5] http://cran.r-project.org/web/packages/raster/
> [6] http://cran.r-project.org/web/packages/spgrass6/
>

-- 
Roger Bivand
Department of Economics, NHH Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no