C R wrote:
> Hi > > Thanks for your reply. Could you just clarify this for me: > > In: > > X = g*w*I + (1-g)*W > > Did you mean: > > X = g*w*I + (1-g)*V > > where V is the original covariance matrix ?? > > Cheers > > "Tony T. Warnock" <[EMAIL PROTECTED]> wrote in message > [EMAIL PROTECTED]">news:[EMAIL PROTECTED]... > > I have used the following based on "regularized discriminant analysis" > > (and I can't remember the author.) > > > > If your matrix is V, then let w=Trace(V) and modify V by a fudge factor > > g (I use g in the range .02 to .05) with: > > > > X = g*w*I + (1-g)*W > > > > The idea is to move the matrix closer to the unit matrix but keep the > > trace the same. > > Yes. Finger failer. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
