[6309] in s-news-athena
Re: g-inv and zeros
daemon@ATHENA.MIT.EDU (Pierre Duchesne)
Mon Jan 30 11:11:28 1995
From: duchesne@STAT.UMontreal.CA (Pierre Duchesne)
To: mcintosh@bellcore.com (Allen Mcintosh)
Date: Mon, 30 Jan 1995 10:21:16 -0500 (EST)
Cc: S-news@utstat.toronto.edu
In-Reply-To: <199501301445.JAA16101@larch.bellcore.com> from "Allen Mcintosh" at Jan 30, 95 09:45:22 am
>
> Why do you want to find the Moore-Penrose inverse, and what are you finding
> the Moore-Penrose inverse of?
>
> If the matrix has the form X'X, you are much better off working with X if at
> all possible. Computing X'X squares the condition number (the ratio of the
> largest to the smallest non-zero eigenvalue). The condition number is, in
> turn, a measure of how "well behaved" the matrix is.
>
You're right, I want to find the g-inv of a matrix of the form X'X in
a projector ( X%*%g.inv(X'X)%*%t(X) ).
Thanks for your good idea to use X not X'X for estimating the rank of X'X...
Bye
Pierre
