gsl_stats_wcovariance

[James Bergstra <james dot bergstra at umontreal dot ca>]

This email concerns the functions

- gsl_stats_wcovariance
- gsl_stats_wcovariance_w

The file wcovar_source.c was in the cvs tree, but was not included in the build
and it appeared to be incomplete.  Under the assumption that it had simply been
forgotten (copyright notice was 2000!), I wrote it up.

I'd appreciate it if someone who knows about unbiased estimators of weighted
sample statistics would have a glance at my algorithm.  I haven't a clue-- I
implemented a straightforward extension of the wvariance routines.

Please find attached (if the mailing list supports attachments) the files that I
modified.  They are from the statistics subdirectory of today's cvs.  In
addition to rewriting wcovar_source.c, I added wcovariance declarations to
headers (double, float, long double) and added wcovar.c to Makefile.am.  Also
attached is a very simple test program.  This program's output agrees with the
cov.wt routine from R.

If anyone encourages me, I'll try to figure out where/how to add the necessary
documentation.  But before I go further, I'd be curious to know why the original
implementation was abandoned.

-- 
James Bergstra
http://www-etud.iro.umontreal.ca/~bergstrj

Attachment: gsl_statistics_double.h
Description: Text document

Attachment: gsl_statistics_long_double.h
Description: Text document

Attachment: gsl_statistics_float.h
Description: Text document

Attachment: wcovar_source.c
Description: Text document

Attachment: Makefile.am
Description: Text document

Attachment: test_wcov.c
Description: Text document