[RsR] estimators based on random samples... - should be random
Martin Maechler
m@ech|er @end|ng |rom @t@t@m@th@ethz@ch
Wed May 10 09:27:02 CEST 2006
>>>>> "ValenT" == Valentin Todorov <valentin.todorov using chello.at>
>>>>> on Wed, 10 May 2006 00:52:01 +0200 writes:
>> By the way, Valentin,
>> do you feel it is important to remain with Fortran there?
>> In general, I end up 'f2c'ing (Fortran to C translating) Fortran
>> source and work with C; not the least because I get better
>> compiler warnings and easier debugging tools.
ValenT> I am not happy with the Fortran in covMcd and ltsReg, but:
ValenT> 1. I still do not believe in the automatic coversion (yes, you are right, I
ValenT> had first at least to try it)
ValenT> 2. I would like to redesign the programs
ValenT> 3. I would prefer C++ rather than C.
C++ complicates things slightly and typically is a tad (10% ?) slower than C.
R's C API is a C and not C++ API anyway.
I think it's only worth using C++ for some kind of problems
(e.g. when a class structure can make the program much more readable),
and I doubt that the covMcd() or ltsReg() algorithms belong to
this class.
ValenT> For all these I still have not found enough time.
ValenT> By the way my trials to download an f2c converter ended with the download of
ValenT> a "Free download Manager"! and nothing more. I'll appreciate some info.
f2c is from the Bell Labs -- AFAIK by some of the same people
who invented C. Good Linux distributions provide a package for
it. Googling 'f2c' gives a good site (http://www.netlib/org/f2c)
as first hit for me.
I've used it -- plus an extra perl script I wrote myself --
many times in the past ten years; even parts of R itself were
done using it.
BTW, I thought of doing the conversion (for fast MCD & LTS)
myself, rather than you having to do it.
Martin
More information about the R-SIG-Robust
mailing list