[算法] 模型选择 Model Selection

2018-11-25  本文已影响0人  数据麻瓜

ESL "模型选择"章的中文小结
https://cosx.org/2015/08/some-basic-ideas-and-methods-of-model-selection/

内容补充:

  1. 符号表示汇总:

由于Err_\tau要引入新的X_0,Y_0有难度,那么退一步看
X 不变动而仅引入新的Y_0的预测误差

  1. Bias-Variance Decomposition:
    Err(x_0) = E[(Y-\hat{f}(x_0))^2|X=x_0]
    \hat{f}(x_0)=\hat{f}, f(x_0)=f
    \begin{aligned} E[ (Y - \hat f)^2 ] &= E[(f + \epsilon - \hat f )^2] \\ & = E[\epsilon^2] + E[(f - \hat f)^2] + 2 E[(f - \hat f)\epsilon] \\ & = (E[\epsilon])^2+Var(\epsilon) + E[(f - \hat f)^2] + 2E[(f−\hat{f})ϵ] \\ & = \sigma^2_\epsilon + E[(f - \hat f)^2] + 0 \end{aligned}

\begin{aligned} E[(f - \hat f)^2] & = E[(f + E[\hat f] - E[\hat f] - \hat f)^2] \\ & = E \left[ f - E[\hat f] \right]^2 + E\left[ \hat f - E[ \hat f] \right]^2 \\ & = \left[ f - E[\hat f] \right]^2 + E\left[ \hat f - E[ \hat f] \right]^2 \\ & = Bias^2[\hat f] + Var[\hat f] \end{aligned}

  1. (2.2.2上方公式的错误)\sum^N_{i=1}\text{Cov}(y_i, \hat{y}_i) = \sum^N_{i=1}\text{Cov}(y_i, \mathbf{Sy}_i) = \text{trace}(\mathbf{S})\sigma_{\epsilon}^2 = d \sigma_{\epsilon}^2

  2. AIC, BIC中的loglik是将MLE代入log likelyhood方程的结果

  3. 解析法和CV之类方法的区别:解析法仅限于线性方法,非解析法跟通用一点,因为他直接估计了extra-sample error,可以适用于任何loss function

  4. CV-一般选5/10折

  5. 正确的进行cross validation的步骤:

7. 常用模型选择方法
上一篇下一篇

猜你喜欢

热点阅读