标签:应用 协方差矩阵 独立 lin splay play 概念 end 规范
在多元变量分析中,我们考虑所有的 \(d\) 个数值型属性 \(X_1, \cdots, X_d\)。整个数据集是一个 \(n \times d\) 的矩阵,即(数据矩阵):
\[ D = \left[ \begin{array}{c|llll} & X_1 & X_2 & \cdots & X_d \ \hline x_1^T & x_{11} & x_{12} & \cdots & x_{1d} \ x_2^T & x_{21} & x_{22} & \cdots & x_{2d} \ \vdots & \vdots & \vdots & \ddots & \vdots \ x_n^T & x_{n1} & x_{n2} & \cdots & x_{nd} \ \end{array} \right] \]
以上数据:
从概率的角度,\(d\) 个属性可以建模为一个向量随机变量 \(X = (X_1, X_2, \cdots, X_d)^T\),而点 \(x_i\) 可以看成从 \(X\) 中得到的随机样本,它们和 \(X\) 是独立同分布的。
\[ \begin{align} \mu = E[X] = \left[ \begin{array}{c} E[X_1] \\ E[X_2] \\ \vdots \\ E[X_d] \end{array} \right] = \left[ \begin{array}{c} \mu_1 \\ \mu_2 \\ \vdots \\ \mu_d \end{array} \right] \tag{均值向量} \\hat{\mu} = \frac{1}{n} \sum_{i=1}^{n}x_i \tag {样本均值} \end{align} \]
\[ \Sigma = E[(X - \mu)(X - \mu)^T] \]
\[ Z = D - 1 \cdot \hat{\mu}^T \]
\[ \hat{\Sigma} = E[(X - \hat{\mu})(X - \hat{\mu})^T] = \frac{1}{n - 1}\; (Z^TZ) \]
\[ var(D) = tr(\Sigma) \]
极差:\(\hat{r} = \max\{X_i\} - \min\{X_i\}\)
\(X_i^{‘} = \frac{X_i - \min\{X_i\}}{\hat{r}}\)
\[ \hat{X} = \frac{X - \hat{\mu}}{\hat{\sigma}} \]
\[ erf(x) = \frac{2}{\sqrt{\pi}}\;\int_0^xe^{-t^2}{\rm d}t \]
随机变量 \(X\) 服从正态分布,均值为 \(\mu\),方差为 \(\sigma^2\),其概率密度函数可以描述为:
\[ f(x\,|\,\mu, \sigma^2) = \frac{1}{\sqrt{2\pi \sigma^2}} \exp\left\{-\frac{(x - \mu)^2}{2 \sigma^2}\right\} \]
给定区间 \([a, b]\),在该区间上的正态分布的概率质量为:
\[ P(a \leq x \leq b) = \int_a^b f(x\,|\,\mu, \sigma^2) {\rm d} x \]
我们大都对于区间 \([\mu - k \sigma, \mu + k \sigma]\) 比较感兴趣:
\[ P(\mu - k \sigma \leq x \leq \mu + k \sigma) = \int_{\mu - k \sigma}^{\mu + k \sigma} f(x\,|\,\mu, \sigma^2) {\rm d} x \]
我们令 \(z = \frac{x - \mu}{\sigma}\),则上式可以化为:
\[ \begin{align} P(- k \leq z \leq k) &= \frac{1}{\sqrt{2\pi}} \int_{- k}^{k} e^ {- \frac{1}{2}{z^2}} {\rm d}z \ &= \frac{2}{\sqrt{2\pi}} \int_{0}^{k} e^ {- \frac{1}{2}{z^2}} {\rm d}z \ &= \frac{2}{\sqrt{\pi}} \int_{0}^{\frac{k}{\sqrt{2}}} e^{- t^2} {\rm d}t \ &= erf(\frac{k}{\sqrt{2}}) \end{align} \]
若 \(X = (X_1, X_2, \cdots, X_d)\) 服从多元正态分布,均值为 \(\bf \mu\),协方差矩阵为 \(\bf \Sigma\),则其联合多元概率密度函数为:
\[ f(x\,|\,\mu, \Sigma) = \frac{1}{\sqrt{2\pi}^d {\sqrt{|{\Sigma}|}}} \exp\left\{-\frac{(x - \mu)^T{\Sigma}^{-1}(x - \mu)}{2} \right\} \]
\[ (x - \mu)^T{\Sigma}^{-1}(x - \mu) \]
标签:应用 协方差矩阵 独立 lin splay play 概念 end 规范
原文地址:https://www.cnblogs.com/q735613050/p/9344606.html