statistical pattern recognition

作者: Rooooooooong | 来源:发表于2018-10-14 15:06 被阅读1次

K是协方差矩阵,X是n*m的矩阵(n纬特征)
X = (X_1,X_2,...X_i,...,X_m)
\Lambda_p = diag(\Lambda_1,\Lambda_2,...,\Lambda_p)\tag{1}
V_p = (V_1,V_2,...,V_p)\tag{2}
XX^T求出式(1)(2),这里同PCA原理一致。差异在式(3):
\overline{X_i} = \Lambda_p^{-\frac{1}{2}}V_p^T(X_i-\mu) \tag{3}
以笔者拙见,这里可以看成对X_i进行了标准化。所以$\overline{X}的协方差矩阵如下:

\begin{align} \overline{X_i}\overline{X_i}^T &= \Lambda_p^{-\frac{1}{2}}V_p^T(X_i-\mu) (X_i-\mu)^TV_p{\Lambda_p^{-\frac{1}{2}}}^T\\ &= \Lambda_p^{-\frac{1}{2}}V_p^TKV_p{\Lambda_p^{-\frac{1}{2}}}^T\\ &= I_p \end{align}\tag{4}
所以\overline{X_i}\sim MVN(0,I_p)
\overline{X_i} = (\overline{x}_{i1},\overline{x}_{i2},...,\overline{x}_{ij},...,\overline{x}_{ip})^T\tag{5}
\overline{X_i}^T\overline{X_i} = \sum_{j=1}^p{\overline{x_{ij}}^2}\tag{6}
CV(\overline{X_i}) = 1-F(\overline{X_i}^T\overline{X_i})\tag{7}
以上,就是核心算法。说白了,statistical pattern recognition 就是PCA基础上嫁接一个
\chi方分布的cdf,用来计算健康分。

相关文章

网友评论

    本文标题:statistical pattern recognition

    本文链接:https://www.haomeiwen.com/subject/iaizaftx.html