Things no longer on M.M's TODO list for R

Here are things that earlier where on my TODO list with the reasons why they are there no longer ...
Martin Maechler maechler@r-project.org $Date: 2004/05/15 13:52:29 $ UTC

Things done by myself (at least partly)

Item	How and when done
Running Medians for library(modreg) : We need a fast robust smoother; `lowess()` is not robust.	done for R 1.7 "unstable", in Feb.2003.
scatterplot3d from Uwe Ligges): A simple 3D scatterplot with somewhat orthogonal functionality to persp()	The package has has been well available from CRAN, is mature and documented, and we have submitted a paper (as of end 2002). In the longer term, we will probably see real 3D graphics much more widely available from R.
Implement C.Loader's dbinom(), dpois() improvements	done for R 1.2 "unstable", as of Sept.29, 2000, together with more, using CL's bd0() function
Replace `qnorm()`'s algorithm with AS 241	done for R 1.1.0 (June 15, 2000)
`cor(.)` should get a new `method` argument, with a `match.arg' default of `c("pearson", "kendall", "spearman")` completely in line with the `cor.test(.)` function in the `ctest` package --- need to move C code for Kendall from ctest to base (and coordinate all this with Kurt!).	done for R 1.8.0 (Oct 2003)
Implement the final Newton step in `qgamma()`'s C code, suggested Morten Welinder, in bug report PR #2214	done for R 1.9.0 (early 2004)

Things done by others

Item	How and when done
Numerical Integration at least as good as Quadpack (see also GSL -- The GNU Scientific Library )	R function integrate() since R 1.3.0 (2001-06-22); C API to same code, for R 1.5.0 (2002-04-26)

Things no longer ``interesting'' to me

Topic Reasoning

Topic	Reasoning
An option which would allow to use the old (quite less precise, but a tad faster) version of `pnorm( )`. Not a new argument to qnorm(), but rather a global new entry in `options(..)`, something like options(prob.precision = 1e-10) options(dpq.precision = 1e-10) (where 1e-10 is somewhat arbitrary) which would use faster less precise algorithms when the required precision was set high (i.e., 1e-4 instead of 1e-10).	Got no feedback on the proposition; The (considerable) drawback would have been that a global option could change numerical results (not just the printing of them), which should only ever happen if really deemed important.

An option which would allow to use the old (quite less precise, but a tad faster) version of pnorm( ). Not a new argument to qnorm(), but rather a global new entry in options(..), something like

      options(prob.precision = 1e-10)
      options(dpq.precision  = 1e-10)

(where 1e-10 is somewhat arbitrary) which would use faster less precise algorithms when the required precision was set high (i.e., 1e-4 instead of 1e-10).

Got no feedback on the proposition; The (considerable) drawback would have been that a global option could change numerical results (not just the printing of them), which should only ever happen if really deemed important.