1.方差

  • 方差是用来衡量一组数据的离散程度,数序表达式如下:
    σ2=1N∑i=1N(xi−μ)2 \sigma^2=\frac1N\sum_{i=1}^N(x_i-\mu)^2 σ2=N1​i=1∑N​(xi​−μ)2

    • σ2σ^2σ2表示样本的总体方差,
    • NNN 表示样本总数,
    • xix _ixi​是第 i 个样本,
    • μμμ 是数据集的平均值(μ=1N∑i=1Nxi)(\mu=\frac1N\sum_{i=1}^Nx_i)(μ=N1​∑i=1N​xi​)
  • 假设某班级有 5 名学生的数学成绩为:85,74,63,95,9985, 74, 63, 95, 9985,74,63,95,99, 计算平均值为 μ=85+74+63+95+995=83.2μ = \frac{85 +74 + 63 + 95 + 99} 5 =83.2μ=585+74+63+95+99​=83.2 ,方差的计算如下:
    σ成绩2=(85−83.5)2+(74−83.5)2+(63−83.5)2+(95−83.5)2+(99−83.5)25=177.05 {σ_{成绩}}^2 = \frac{(85-83.5)^2 +(74-83.5)^2 + (63-83.5)^2 + (95-83.5)^2 + (99 - 83.5)^2}{5} = 177.05 σ成绩​2=5(85−83.5)2+(74−83.5)2+(63−83.5)2+(95−83.5)2+(99−83.5)2​=177.05

2.协方差

  • 协方差是用来衡量两组样本之间的关系,数序表达式如下:
    σxσy=1N∑i=1N(xi−μx)(yi−μy) \sigma_x\sigma_y=\frac{1}{N}\sum_{i=1}^N(x_i-\mu_x)(y_i-\mu_y) σx​σy​=N1​i=1∑N​(xi​−μx​)(yi​−μy​)
    • σxσyσ_xσ_yσx​σy​表示两组样本的总体协方差,
    • NNN 是样本对的总数,
    • xi,yix _i, y_ixi​,yi​ 是第 i 对样本,
    • μx,μyμ_x,μ_yμx​,μy​分别表示两组样本的均值。
  • 例如,要计算如下 x,yx, yx,y 两个样本的协方差:
xy
500200
600250
451180
    • 计算xxx的平均值 ux=500+600+4513=517u_x = \frac{500 +600 +451}3 = 517ux​=3500+600+451​=517
    • 计算yyy的平均值 uy=200+250+1803=210u_y = \frac{200 +250 +180}3 = 210uy​=3200+250+180​=210
      σxσy=(500−517)(200−210)+(600−517)(250−210)+(451−517)(180−210)3=1823.33... σ_xσ_y = \frac{(500-517)(200 - 210) + (600-517)(250-210) + (451-517)(180-210) }{3} = 1823.33... σx​σy​=3(500−517)(200−210)+(600−517)(250−210)+(451−517)(180−210)​=1823.33...

3.协方差矩阵

协方差矩阵就是将多组数据的方差和协方差用矩阵的形式表达出来,例如如下三组样本,其协方差的矩阵排布如下,从下图可以看出,对角线上是各个样本的方差,两边则是样本之间的协方差

xyz
x1x_1x1​y1y_1y1​z1z_1z1​
x2x_2x2​y2y_2y2​z2z_2z2​
x3x_3x3​y3y_3y3​z3z_3z3​

cov=[σx2σxσyσxσzσyσxσy2σyσzσzσxσzσyσx2] cov = \left[\begin{array}{ccc}σ_x^2&σ_xσ_y&σ_xσ_z\\ σ_yσ_x&σ_y^2&σ_yσ_z\\ σ_zσ_x&σ_zσ_y&σ_x^2\end{array}\right] cov=​σx2​σy​σx​σz​σx​​σx​σy​σy2​σz​σy​​σx​σz​σy​σz​σx2​​​
在这里插入图片描述

4.通过矩阵运算求解一个矩阵的协方差矩阵

  • 将第三节使用的样本写成矩阵的形式,本节主要求解这个矩阵的协方差矩阵
    [x1y1z1x2y2z2x3y3z3] \left[\begin{array}{ccc}x_1&y_1&z_1\\ x_2&y_2&z_2\\ x_3&y_3&z_3\end{array}\right] ​x1​x2​x3​​y1​y2​y3​​z1​z2​z3​​​
  • 求出过度矩阵 aaa,根据下面计算可知,过度矩阵 aaa 其实是求解了每个样本的 标准差
    a=[x1y1z1x2y2z2x3y3z3]−13[111111111][x1y1z1x2y2z2x3y3z3]−−−−a=[x1y1z1x2y2z2x3y3z3]−13[x1+x2+x3y1+y2+y3z1+z2+z3x1+x2+x3y1+y2+y3z1+z2+z3x1+x2+x3y1+y2+y3z1+z2+z3]−−−−a=[x1−13(x1+x2+x3)y1−13(y1+y2+y3)z1−13(z1+z2+z3)x2−13(x1+x2+x3)y2−13(y1+y2+y3)z2−13(z1+z2+z3)x3−13(x1+x2+x3)y3−13(y1+y2+y3)z3−13(z1+z2+z3)] a = \left[\begin{array}{ccc}x_1&y_1&z_1\\x_2&y_2&z_2\\x_3&y_3&z_3\end{array}\right] - \frac{1}{3} \left[\begin{array}{ccc}1&1&1\\1&1&1\\1&1&1\end{array}\right] \left[\begin{array}{ccc}x_1&y_1&z_1\\x_2&y_2&z_2\\x_3&y_3&z_3\end{array}\right] \\ ----\\ a = \left[\begin{array}{ccc}x_1&y_1&z_1\\x_2&y_2&z_2\\x_3&y_3&z_3\end{array}\right] - \frac{1}{3} \left[\begin{array}{ccc}x_1+x_2+x_3&y_1+y_2+y_3&z_1+z_2+z_3\\x_1+x_2+x_3&y_1+y_2+y_3&z_1+z_2+z_3\\x_1+x_2+x_3&y_1+y_2+y_3&z_1+z_2+z_3\end{array}\right] \\ ---- \\ a = \left[\begin{array}{ccc}x_1 - \frac{1}{3} (x_1+x_2+x_3)&y_1 - \frac{1}{3} (y_1+y_2+y_3)&z_1 -\frac{1}{3} (z_1+z_2+z_3) \\x_2 - \frac{1}{3} (x_1+x_2+x_3)&y_2 - \frac{1}{3} (y_1+y_2+y_3)&z_2 -\frac{1}{3} (z_1+z_2+z_3)\\x_3 - \frac{1}{3} (x_1+x_2+x_3)&y_3 - \frac{1}{3} (y_1+y_2+y_3)&z_3 -\frac{1}{3} (z_1+z_2+z_3)\end{array}\right] a=​x1​x2​x3​​y1​y2​y3​​z1​z2​z3​​​−31​​111​111​111​​​x1​x2​x3​​y1​y2​y3​​z1​z2​z3​​​−−−−a=​x1​x2​x3​​y1​y2​y3​​z1​z2​z3​​​−31​​x1​+x2​+x3​x1​+x2​+x3​x1​+x2​+x3​​y1​+y2​+y3​y1​+y2​+y3​y1​+y2​+y3​​z1​+z2​+z3​z1​+z2​+z3​z1​+z2​+z3​​​−−−−a=​x1​−31​(x1​+x2​+x3​)x2​−31​(x1​+x2​+x3​)x3​−31​(x1​+x2​+x3​)​y1​−31​(y1​+y2​+y3​)y2​−31​(y1​+y2​+y3​)y3​−31​(y1​+y2​+y3​)​z1​−31​(z1​+z2​+z3​)z2​−31​(z1​+z2​+z3​)z3​−31​(z1​+z2​+z3​)​​
  • 最后求解协方差矩阵P
    P=13∗aT∗a P = \frac{1}{3} * a^T *a P=31​∗aT∗a

aT=[x1−13(x1+x2+x3)x2−13(x1+x2+x3)x3−13(x1+x2+x3)y1−13(y1+y2+y3)y2−13(y1+y2+y3)y3−13(y1+y2+y3)z1−13(z1+z2+z3)z2−13(z1+z2+z3)z3−13(z1+z2+z3)] a^T = \left[\begin{array}{ccc}x_1 - \frac{1}{3} (x_1+x_2+x_3)&x_2 - \frac{1}{3} (x_1+x_2+x_3)&x_3 - \frac{1}{3} (x_1+x_2+x_3) \\ y_1 - \frac{1}{3} (y_1+y_2+y_3)&y_2 - \frac{1}{3} (y_1+y_2+y_3)&y_3 - \frac{1}{3} (y_1+y_2+y_3)\\ z_1 - \frac{1}{3} (z_1+z_2+z_3)&z_2 -\frac{1}{3} (z_1+z_2+z_3)&z_3 -\frac{1}{3}(z_1+z_2+z_3)\end{array}\right] aT=​x1​−31​(x1​+x2​+x3​)y1​−31​(y1​+y2​+y3​)z1​−31​(z1​+z2​+z3​)​x2​−31​(x1​+x2​+x3​)y2​−31​(y1​+y2​+y3​)z2​−31​(z1​+z2​+z3​)​x3​−31​(x1​+x2​+x3​)y3​−31​(y1​+y2​+y3​)z3​−31​(z1​+z2​+z3​)​​

P=13∗[x1−13(x1+x2+x3)x2−13(x1+x2+x3)x3−13(x1+x2+x3)y1−13(y1+y2+y3)y2−13(y1+y2+y3)y3−13(y1+y2+y3)z1−13(z1+z2+z3)z2−13(z1+z2+z3)z3−13∗(z1+z2+z3)]∗[x1−13(x1+x2+x3)y1−13(y1+y2+y3)z1−13(z1+z2+z3)x2−13(x1+x2+x3)y2−13(y1+y2+y3)z2−13(z1+z2+z3)x3−13(x1+x2+x3)y3−13(y1+y2+y3)z3−13(z1+z2+z3)] P = \frac{1}{3} * \left[\begin{array}{ccc}x_1 - \frac{1}{3} (x_1+x_2+x_3)&x_2 - \frac{1}{3} (x_1+x_2+x_3)&x_3 - \frac{1}{3} (x_1+x_2+x_3) \\ y_1 - \frac{1}{3} (y_1+y_2+y_3)&y_2 - \frac{1}{3} (y_1+y_2+y_3)&y_3 - \frac{1}{3} (y_1+y_2+y_3)\\ z_1 - \frac{1}{3} (z_1+z_2+z_3)&z_2 -\frac{1}{3} (z_1+z_2+z_3)&z_3 -\frac{1}{3} *(z_1+z_2+z_3)\end{array}\right] * \left[\begin{array}{ccc}x_1 - \frac{1}{3} (x_1+x_2+x_3)&y_1 - \frac{1}{3} (y_1+y_2+y_3)&z_1 -\frac{1}{3} (z_1+z_2+z_3) \\x_2 - \frac{1}{3} (x_1+x_2+x_3)&y_2 - \frac{1}{3} (y_1+y_2+y_3)&z_2 -\frac{1}{3} (z_1+z_2+z_3)\\x_3 - \frac{1}{3} (x_1+x_2+x_3)&y_3 - \frac{1}{3} (y_1+y_2+y_3)&z_3 -\frac{1}{3} (z_1+z_2+z_3)\end{array}\right] P=31​∗​x1​−31​(x1​+x2​+x3​)y1​−31​(y1​+y2​+y3​)z1​−31​(z1​+z2​+z3​)​x2​−31​(x1​+x2​+x3​)y2​−31​(y1​+y2​+y3​)z2​−31​(z1​+z2​+z3​)​x3​−31​(x1​+x2​+x3​)y3​−31​(y1​+y2​+y3​)z3​−31​∗(z1​+z2​+z3​)​​∗​x1​−31​(x1​+x2​+x3​)x2​−31​(x1​+x2​+x3​)x3​−31​(x1​+x2​+x3​)​y1​−31​(y1​+y2​+y3​)y2​−31​(y1​+y2​+y3​)y3​−31​(y1​+y2​+y3​)​z1​−31​(z1​+z2​+z3​)z2​−31​(z1​+z2​+z3​)z3​−31​(z1​+z2​+z3​)​​

  • 如此运算便可通过矩阵运算求出协方差矩阵
Logo

有“AI”的1024 = 2048,欢迎大家加入2048 AI社区

更多推荐