Re: regression.q

[John Darrington]

Thanks for the feedback.  I checked this in, and will give it more
thorough testing over the next few days.

Like you say, it would be good if we could reconcile glm and regression.

J'

On Thu, Apr 12, 2012 at 05:43:06PM -0400, Jason Stover wrote:
     On Thu, Apr 12, 2012 at 07:43:54PM +0000, John Darrington wrote:
     > There are a couple of open bugs agains regression.
     > 
     > I think it'll be easier to maintain this code if we move away from the
     > .q format.  Hence, I'm proposing the attached .c as a drop in 
     > replacement.   Most of it I just copied from the existing .q
     > 
     > Comments?
     
     I know I have been silent for a while, 'sorry. Busy.
     
     I scanned the file and it looks OK to me.
     
     My main comment is that, given the bugs and the fact that glm.c is now
     working, it may be worth dropping linreg.c sometime in the future and
     instead using the same approach as glm.c: Compute a covariance matrix,
     then call code in sweep.c directly, and add support for gsl's
     least-squares module.  (The former uses the sweep operator, the latter
     uses QR decomposition, both of which are supported in the other
     software.)  I think such an approach would make REGRESSION easier to
     maintain, and get rid of the clunkier linreg.c.

-- 
PGP Public key ID: 1024D/2DE827B3 
fingerprint = 8797 A26D 0854 2EAB 0285  A290 8A67 719C 2DE8 27B3
See http://keys.gnupg.net or any PGP keyserver for public key.

signature.asc
Description: Digital signature