随机变量的相似性度量
随机变量是如何度量相似性/相关性的?
方差 D ( x i ) = c o v ( x i , x i ) = E 2 ( x i − E ( x i ) ) D(x_i)=cov(x_i,x_i)=E^2(x_i-E(x_i)) D(xi)=cov(xi,xi)=E2(xi−E(xi))
事实上,对 x \mathbf x x而言,其方差是一个矩阵,记作协差阵 Σ \Sigma Σ,即 Σ = [ c o v ( x 1 , x 1 ) ⋯ c o v ( x 1 , x p ) ⋮ ⋮ c o v ( x p , x 1 ) ⋯ c o v ( x p , x p ) ] = [ σ 11 ⋯ σ 1 p ⋮ ⋮ σ p 1 ⋯ σ p p ] \Sigma=\begin{bmatrix}cov(x_1,x_1)&\cdots&cov(x_1,x_p)\\\vdots&&\vdots \\cov(x_p,x_1)&\cdots&cov(x_p,x_p)\end{bmatrix}=\begin{bmatrix}\sigma_{11}&\cdots&\sigma_{1p}\\\vdots&&\vdots \\\sigma_{p1}&\cdots&\sigma_{pp}\end{bmatrix} Σ=⎣⎢⎡cov(x1,x1)⋮cov(xp,x1)⋯⋯cov(x1,xp)⋮cov(xp,xp)⎦⎥⎤=⎣⎢⎡σ11⋮σp1⋯⋯σ1p⋮σpp⎦⎥⎤
多维随机变量的相似性由相关系数阵 R \mathbf R R度量 R = [ c o v ( x i , x j ) σ i i σ j j ] = [ r 11 ⋯ r 1 p ⋮ ⋮ r p 1 ⋯ r p p ] \mathbf R=\begin{bmatrix}\frac{cov(x_i,x_j)}{\sqrt{\sigma_{ii}}\sqrt{\sigma_{jj}}}\end{bmatrix}=\begin{bmatrix}r_{11}&\cdots&r_{1p}\\\vdots&&\vdots \\r_{p1}&\cdots&r_{pp}\end{bmatrix} R=[σiiσjj