#klgauss.m#

来自「Kalman滤波开发包（matlab版）包括很多编得很好得函数」· M# 代码 · 共 11 行

function kl = KLgauss(P, Q)
%The following computes D(P||Q), the KL divergence between two zero-mean
%Gaussians with covariance P and Q:
%  klDiv = -0.5*(log(det(P*inv(Q))) + trace(eye(N)-P*inv(Q)));

R = P*inv(Q);
kl = -0.5*(log(det(R))) + trace(eye(length(P))-R);

To get MI, just set P=cov(X,Y) and Q=blockdiag(cov(X),cov(Y)).  The formula
with differing means is not much more complicated, but I don't have code for it
and can't remember the formula at the moment.