R devel - Mar 2004 - R_DT_val accuracy (PR#6692)

Full_Name: M. Welinder
Version: 1.8.1
OS: Solaris
Submission from: (NULL) (65.213.85.227)


Currently R has...

#define R_D_Lval(p)	(lower_tail ? (p) : (1 - (p)))	/*  p  */
#define R_D_val(x)	(log_p	? log(x) : (x))		/*  x  in pF(x,..) */
#define R_DT_val(x)	R_D_val(R_D_Lval(x))		/*  x  in pF */

...which is sub-optimal in the lower_tail==FALSE && log_p==TRUE case.
Something like this ought to work better.

#define R_DT_val(x)	(log_p	? (lower_tail ? log(x) : log1p (-(x))) :
(R_D_Lval(x)))

Similarly for R_DT_Cval.

This affects the accuracy of phyper, for example.

>>>>> "Morten" == Morten Welinder
<terra@gnome.org>
>>>>>     on Wed, 24 Mar 2004 15:53:21 +0100 (CET) writes:
    Morten> Full_Name: M. Welinder Version: 1.8.1 OS: Solaris
    Morten> Submission from: (NULL) (65.213.85.227)


    Morten> Currently R has...
            ^^^^^^^^^
not quite true!

Currently, R is in beta testing phase,
and everyone is

     *really* *really* *really* 

kindly asked to work with R 1.9.0 beta  -- particularly before
even considering sending a bug report. 

I've done something along your suggestion quite a few weeks ago
and the NEWS entry has been for a while

    o	The internal header nmath/dpq.h has slightly improved macros
	R_DT_val() and R_DT_Cval(), a new R_D_LExp() and improved
	R_DT_log() and R_DT_Clog();  this improves accuracy in several
	[dpq]-functions {for "extreme" arguments}.


Regards,
Martin Maechler, 
Seminar fuer Statistik, ETH Zurich, Switzerland