Martin Maechler's TODO list for R
This page contains (some of) my plans for enhancements to
R ... with absolutely no warranty ...
You might also be interested in things
no longer on the current list below.
Martin Maechler
maechler@r-project.org
Shorter term
- Improve pair() to allow spreading over a few pages
("wall paper" :-) {a bit as lag.plot() (from the
ts package) does}.
Improve it along the lines of Catherine Hurley (even much easier)
- plot.factor() should *not* do boxplots by default when
the boxplot contains less than or only about 5-7 observations!
- Extend box() to also accept 'bg' and draw backgrounds using GRect().
box(bg="green3") will then be a higher level alternative to
(code from demo(graphics)!)
usr <- par("usr")
rect(usr[1], usr[3], usr[2], usr[4], col="green3")
box()
- The dendrogram class {package "stats"} needs better and more versatile
methods, particularly plot.*().
One application is for the
VLMC package available from CRAN. Here, do
as.dendrogram.VLMC() properly and implement plot.VLMC() basically
doing plot.dendrogram( as.dendrogram.VLMC( * ) ).
(done in parts for R 1.9.0; more todo)
- Minimum Spanning Tree: two versions (maybe in one function):
- mstree() using Euclidean distance and the usual n x p data
matrix. Robert has a simple mstree() in his unofficial
Genecluster package in bioconductor {though lacking Friedman/Rafsky's
planing option} which should be easy to put into 'stats' as is.
- I really want to compute the general MST problem (i.e. given an
arbitrary graph with weighted edges, find a subgraph which is a tree
with minimal edge weight sum). The typical application I have in mind
would still be graphs with many edges, i.e. almost complete ones.
- procrustes() is wanted, but not just the two sample case,
and (eventually at least) also with case and variable weighting.
- Extend the Guidelines for Rd files,
(say more about \item in \arguments,
also recommendation/examples about the usage of \describe)
and think about merging (or moving) parts into the
Writing R Extensions manual which is part of R, or
available from the
CRAN Manuals.
- cov.wt() should also work with NA's, with a `use' argument identical
to cov(), cor();
Using a notion of "generalized ranks", maybe could also grow "spearman"
and "kendall" methods.
- str() for lists (and data.frames) should get an option to allow
numbering the list components (maybe via a smart
specification of the indent.str argument).
Of less priority
- Need C API for the Median() and more generally quantile(x, alpha).
(not sure anymore; we have a C API to partial sorting)
- stem() should exclude outliers and list them separately
(as S does).
- New print option for printing exact "0"s:
Consider using "." instead of "0" in numerical
matrices. This would make sparse matrices much more readable.
[I have been doing this with S-plus for years.]
- scat1d() {improved rug(); the thing I've co-authored and has then
been put into Hmisc}.
Even Longer Term
- Differential Equation Solvers; I need a boundary value problem solver,
additional to the LSODA functionality of the very good
odesolve package available from
CRAN.
Return to the R Developer Page.
Last update at $Date: 2004/06/04 17:42:50 $ UTC (aka GMT)
Martin Mächler ,
Seminar für Statistik, ETH Zurich,
Switzerland.