Re: [eigen] cache-friendly matrix inverse

[ Thread Index | Date Index | More Archives ]

I'm not sure that this is still on-topic, but if you have a model
	theta = M y + epsilon
(theta, y, and epsilon are vectors;  M is a non-square matrix)
	<epsilon epsilon^T> = V
then the least-squares estimate for theta is
	theta = (M^T V^{-1} M)^{-1} M V^{-1} y
with covariance
	(M^T V^{-1} M)^{-1}

So you really need to invert a matrix.  You can invert it any
way you like (e.g. eigen-value decomposition; Gaussian elimination; ...)
but invert it you must.


Mail converted by MHonArc 2.6.19+