Why is the first factor in Kernel Principal Component Analysis a very big number?

1 次查看（过去 30 天）

gsourop 2017-7-7

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/347707-why-is-the-first-factor-in-kernel-principal-component-analysis-a-very-big-number

Hi everyone,

I am using KPCA with the function provided by Ambarish Jash (https://uk.mathworks.com/matlabcentral/fileexchange/27319-kernel-pca). The first component extracted from the matrix is significantly different than the others. Hence, when I try to regress the component with a simple ols I get a message regarding the singularity of the matrix. A sample code could be some thing like:

x = randn(555,10);
  [Kcomponents]=kernelpca(x',10);
  Kcomponents= Kcomponents';
y=randn(555,1);
y= y(2:end);
KCOM= Kcomponents(1:end-1,1:2);
T=size(y,1);
results_K=ols(y,[ones(T,1) KCOM]);

However, if I reduce the dimensions of x from 555x10 to 100x10, I do not get the same message. Does anyone know why this happens? I would appreciate any help. Thanks in advance