图文并茂的PCA教程

\huge \fn_jvn \huge D=\frac{1}{m}YY^{T} =\\\frac{1}{m}(PX)(PX)^{T} =\frac{1}{m}PXX^{T}P^{T}=\\\frac{1}{m}P(XX^{T})P^{T}= P(\frac{1}{m}XX^{T})P^{T}=PCP^{T} \\Y_{k,n}=P_{k,n}X_{m,n}

n为数据个数,m为数据维数,若达到降维,则k<m。

\huge Y=\bigl(\begin{smallmatrix} a_{1}& a_{2}& ...& a_{n}\\ .&.&...&.&\\ .&.&...&.&\\ .&.&...&.&\\ z_{1}&z_{2}&...&z_{n} \end{smallmatrix}\bigr)\\ Y^{T}=\bigl(\begin{smallmatrix} a_{1}& b_{1}& ...& z_{1}\\ .& .& .&.&\\ .&.&.&.&\\ .&.&.&.&\\ a_{n}&b_{n}&...&z_{n} \end{smallmatrix}\bigr)\\ \frac{1}{m}YY^{T}=\frac{1}{m}\bigl(\begin{smallmatrix} \sum a_{i}^{2}& \sum a_{i}b_{i}& \sum a_{i}c_{i}...&\\ \sum b_{i}a_{i}& \sum b_{i}^{2}& ...&\\ \sum c_{i}a{i}& \end{smallmatrix}\bigr)

\huge \frac{1}{m}YY^{T}_{m,m} \\ E^{T}CE=\Lambda =\bigl(\begin{smallmatrix} \lambda _{1} & \\ & \lambda_{2}& \\ & & \lambda_{3} & & \ & & \\ & & & ...&\\ & & & & \lambda_{m} \end{smallmatrix}\bigr)\\ Q=E^{T}\\ QCQ^{T}=\bigl(\begin{smallmatrix} \lambda _{1} & \\ & \lambda_{2}& \\ & & \lambda_{3} & & \ & & \\ & & & ...&\\ & & & & \lambda_{m} \end{smallmatrix}\bigr)\\

PCA目:Y协方差矩阵对角线不为0,其余为0,为了达到降维效果,取前K个方差。

https://blog.csdn.net/hustqb/article/details/78394058

猜你喜欢

转载自blog.csdn.net/u013049912/article/details/84989280
今日推荐