neue Rmetrics packages

[Diethelm Wuertz]

Dirk Eddelbuettel wrote:

First many thanks to Dirk ...
> Following up on the new Rmetrics packages, I am a little dismayed by the
> increased dependencies.  Here is a query from my CRANberries database behind
> my RSS feed of CRAN entries:
>
> sqlite> select package,version,depends from packages where package like 'f%' and version like '260%';
> package,version,depends
> fArma,260.71,"R (>= 2.2.0), fBasics"
> fAsianOptions,260.71,"R (>= 2.2.0), fOptions"
> fBasics,260.71,"R (>= 2.2.0), fImport"
> fBonds,260.71,"R (>= 2.2.0), fBasics"
> fCopulae,260.71,"R (>= 2.2.0), adapt, fBasics, fMultivar"
> fEcofin,260.71,"R (>= 2.2.0), fUtilities"
> fExoticOptions,260.71,"R (>= 2.2.0), fOptions"
> fExtremes,260.71,"R (>= 2.2.0), fBasics"
> fGarch,260.71,"R (>= 2.2.0), fBasics, fArma"
> fImport,260.71,"R (>= 2.2.0), fSeries"
> fMultivar,260.71,"R (>= 2.2.0), methods, sn, fBasics"
> fRegression,260.71,"R (>= 2.2.0), methods, mgcv, nnet, polspline, fTrading, fMultivar"
> fSeries,260.71,"R (>= 2.2.0), robustbase, fCalendar"
> fTrading,260.71,"R (>= 2.2.0), fBasics"
> fUnitRoots,260.71,"R (>= 2.2.0), urca, fBasics"
> fUtilities,260.71,"R (>= 2.2.0), RUnit, methods, MASS, spatial, akima, tcltk, zoo"
>
> and missing from that were
>
> fCalendar|251.70|R (>= 2.2.1), methods, fEcofin
> fOptions|251.70|R (>= 2.2.1), methods, fEcofin, fBasics, fCalendar, fSeries
> fPortfolio|251.70|R (>= 2.2.1), methods, robustbase, MASS, fEcofin, fCalendar, fBasics, fSeries, fMultivar, fExtremes, fCopulae, quadprog, lpSolve
>   
First I have seen, that the latest versions of fCalendar, fOptions, 
fAssets, and fPortfolio are missing,
so I have them resubmitted to the FTP incoming CRAN server.
>
> So trying to put a dependency tree together:
>
> 1. fUtilities,260.71,"R (>= 2.2.0), RUnit, methods, MASS, spatial, akima, tcltk, zoo"
> 	already fails as RUnit and akima are NOT
> 	in Debian, and akima can't for license reasons
> 	we are checkmate at the outset
>
> 2. fEcofin,260.71,"R (>= 2.2.0), fUtilities"
>
> 3. fCalendar|251.70|R (>= 2.2.1), methods, fEcofin
>
> 4. fSeries,260.71,"R (>= 2.2.0), robustbase, fCalendar"
> 	robustbase is NOT in Debian
>
> 5. fImport,260.71,"R (>= 2.2.0), fSeries"
>
> 6. fBasics,260.71,"R (>= 2.2.0), fImport"
>
> 7. fArma,260.71,"R (>= 2.2.0), fBasics"
>    fBonds,260.71,"R (>= 2.2.0), fBasics"
>    fExtremes,260.71,"R (>= 2.2.0), fBasics"
>    fGarch,260.71,"R (>= 2.2.0), fBasics, fArma"
>    fMultivar,260.71,"R (>= 2.2.0), methods, sn, fBasics"
> 	sn is NOT in Debian 
>    fTrading,260.71,"R (>= 2.2.0), fBasics"
>    fUnitRoots,260.71,"R (>= 2.2.0), urca, fBasics"
>    fOptions|251.70|R (>= 2.2.1), methods, fEcofin, fBasics, fCalendar, fSeries
>
> 8. fAsianOptions,260.71,"R (>= 2.2.0), fOptions"
>    fCopulae,260.71,"R (>= 2.2.0), adapt, fBasics, fMultivar"
>    fExoticOptions,260.71,"R (>= 2.2.0), fOptions"
>    fRegression,260.71,"R (>= 2.2.0), methods, mgcv, nnet, polspline, fTrading, fMultivar"
>    fPortfolio|251.70|R (>= 2.2.1), methods, robustbase, MASS, fEcofin, fCalendar, fBasics, fSeries, fMultivar, fExtremes, fCopulae, quadprog, lpSolve
>   

The following order of the packages will give you some guidance on the 
dependence tree:


fUtilities          WARN
fEcofin             OK
fCalendar           ? MISSING
fSeries             OK
fImport             OK
fBasics             OK
fArma               OK
fGarch              OK
fNonlinear          ? MISSING
fUnitRoots          OK
fTrading            OK
fMultivar           OK
fRegression         OK
fExtremes           OK
fCopulae            OK
fOptions            ? MISSING
fExoticOptions      OK
fAsianOptions       OK
fBonds              OK
fAssets             ? MISSING
fPortfolio          ? MISSING

higher packages have only dependences on previous listed packages.

There is also a package named Rmetrics which should load all
packages one after the other as given in the dependence tree.

>
> So if I want to keep Rmetrics in Debian, even if I added RUnit and spatial I
> would get into problems with akima.   Robustbase is probably a worthwhile
> addtion, as is RUnit, esp. as neither depends on anything else. sn depends on
> mnormt, so that would be two more for fMultivar.
>
> The bug issue is akima:
> "License:  Fortran code: ACM, free for non-commercial use, R functions GPL"
>   
This could be solved may in the following way:

The scientific GNU library has an akima interpolation function:

http://www.uni-koeln.de/rrzk/software/mathematik/gsl/gsl-ref-html/gsl-ref_26.html

Is somebody around who can interface the interpolation functions from 
the GNU
scientific library to R? There is an example for special functions in 
the contributed
R package - gsl 
<http://cran.r-project.org/src/contrib/Descriptions/gsl.html> wrapper 
for the Gnu Scientific Library.




There is another problem with the package fPortfolio. It uses currently the
nonlinear solver with nonlinear constraints Rdonlp2 - it comes as a builtin
with the package.

http://arumat.net/Rdonlp2/

I think this has also to be replaced by another GNU or GNU like licensed 
solver.
A possible candidate may be

http://www.coin-or.org/Bonmin/Intro/index.html



Another question - is there a reason why robustbase is not part of Debian ?
Rmetrics only needs the robust mean-cov estimators - Should I possibly
go back and copy this part of the code to fAssets, or can we find another
solution ?


Diethelm



> This would out _all_ of RMetrics outside of Debian's main and into contrib.
>
> Anybody have any good ideas regarding this?
>
> Dirk
>
>