计算指数分布族的矩（统计推断）
假定指数分布族的形式如下：
f ( x ∣ θ ) = h ( x ) c ( θ ) exp ( ∑ i = 1 k w i ( θ ) t i ( x ) ) f(x|\pmb\theta) = h(x)c(\pmb\theta)\exp\left(\sum^k_{i=1}w_i(\pmb\theta)t_i(x) \right) f(x∣θθθ)=h(x)c(θθθ)exp(i=1∑kwi(θθθ)ti(x))
要保证 h ( x ) h(x) h(x)和 c ( θ ) c(\pmb\theta) c(θθθ)恒大于等于0
则，均值和方差计算方式如下：
E ( ∑ i = 1 k ∂ w i ( θ ) ∂ θ j t i ( X ) ) = − ∂ ∂ θ j log c ( θ ) V a r ( ∑ i = 1 k ∂ w i ( θ ) ∂ θ j t i ( X ) ) = − ∂ 2 ∂ θ j 2 log c ( θ ) − E ( ∑ i = 1 k ∂ 2 w i ( θ ) ∂ θ j 2 t i ( X ) ) \begin{aligned}E \left(\sum^k_{i=1}\frac{\partial w_i(\pmb\theta)}{\partial \theta_j}t_i(X) \right) &= -\frac{\partial}{\partial \theta_j}\log c(\pmb\theta) \\ Var\left(\sum^k_{i=1}\frac{\partial w_i(\pmb\theta)}{\partial \theta_j} t_i(X)\right) &=-\frac{\partial^2}{\partial \theta_j^2}\log c(\pmb\theta) -E\left(\sum^k_{i=1}\frac{\partial^2 w_i(\pmb\theta)}{\partial \theta_j^2} t_i(X) \right) \end{aligned} E(i=1∑k∂θj∂wi(θθθ)ti(X))Var(i=1∑k∂θj∂wi(θθθ)ti(X))=−∂θj∂logc(θθθ)=−∂θj2∂2logc(θθθ)−E(i=1∑k∂θj2∂2wi(θθθ)ti(X))
公式看起来很复杂，但是却很好用
1217
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
