| Author | Message | ||
     shankar kumar (Shankar) |
i could find the score and loading for the given NIR data the size are as follows; data=5 x 44 score=44 x 5 load=5 x 5 how to fix my PCs greater than 5 because load matrix is 5 x 5 please any one could find me the answer | ||
| Christopher D. Brown |
Shankar, I'm not sure how kernal PCA relates to your question, since you make no mention of a kernel. As such I'll assume you are referring only to PCA. Five dimensions will describe five samples perfectly (i.e., your data are rank 5), so PCA can only resolve 5 principal components. Any more than 5 and the dimensions would have to be arbitrarily defined. If you require more dimensions for whatever reason, you'll need to collect more than 5 samples. ~ C. | ||
| shankar |
ok i can understand the thing. the relation of kernal is that iam working with NIR data for plastics so my data contains more number of variables than my samples so i go for kernal method | ||
| Christopher D. Brown |
If you are using a kernel, then you've already defined a nonlinear transformation to the feature space, which will also define your kernel principal components. Is it a gaussian kernel? Polynomial? Radial? The kernel principal components will be limited to 44*5, but the linear components will be limited to min(44,5). My (gentle) observation is that you're not sure what you're doing with a kernel. In which case, I'd recommend you don't use one. ~ C. | ||
| Jay Liu (Jayliu) |
Shankar, Help me clear my misunderstanding. What do you mean by saying 'kernel'? Do you mean 'kernel' algorithm for speeding up PCA calculation for wide matrices or 'kernel' PCA as one of nonlinear PCAs using 'kernel' trick in SVM? | ||
| shankar |
lui, thank you u , your question is very clear . iam talking about the Kernal method for speeding up PCA not for any othere purpose. |