[RsR] First step towards integrating robust and robustbase+friends
Kjell Konis
kje||@kon|@ @end|ng |rom ep||@ch
Tue Jul 8 17:58:24 CEST 2008
Dear Robustnik Community,
The original fastmcd code in the Robust Library has been problematic
with newer versions of R and on 64-bit platforms. Instead of trying to
fix it, I decided to remove that code and use the covMcd function in
robustbase. I'm 99% sure I've got everything working the way it should
and I have submitted a revised package 0.3-3 to CRAN. Anyway, I had a
few thoughts while doing this that I think could benefit from some
discussion on the list.
1) The function covMcd in robustbase (and the very similar function
named CovMcd in rrcov) returns a reweighted estimate based on an
initial MCD estimate. In the Robust Library we made the decision to
separate these two phases: in covRob we have estim = "mcd" for an MCD
estimator (scaled for consistency) and estim = "weighted" for the
reweighted estimator. Is it still worthwhile to make this distinction?
2) Following from 1. Is there really a need for both covMcd and
CovMcd? AFAICT they both do pretty much the same thing except that
CovMcd returns an S4 class rather than a list. From a user's point of
view, which is to be preferred?
3) What do people think about the Robust Library depending on
functions in other packages? Is anyone bothered that this in some way
masks their contribution?
4) The goal of the Robust Library should be to a) provide usable
robust methods, b) continue to work with the examples in MMY (although
output may now differ from that published), and c) advertise functions
in other packages (for instance covRob using covMcd and covMest).
Thank you for your thoughts.
Kjell
More information about the R-SIG-Robust
mailing list