[Rd] pbeta accuracy and efficiency (PR#7153)

[terra at gnome.org]

Full_Name: M Welinder
Version: 1.9
OS: Solaris
Submission from: (NULL) (65.213.85.218)


pbeta has known accuracy and efficiency problems as noted in the code.

Gnumeric 1.3.1 is now out and features a new implementation of pbeta (with R's
calling conventions).  The code is thanks to Ian Smith and lives in
src/mathfunc.c
I know of no problematic arguments.  Gnumeric uses "gnm_float" where R uses
"double",
but this should all be easily fixable with a little search-and-replace.

This should facilitate improved qbeta, pt, pbinom [bug 5727], pnbinom, pf,
and whatever those drag with them.   Slight care will be needed to avoid
squeezing
pbeta's first argument up against 1, see Gnumeric's pf function, for example.