I have a training data X_centered of size 120 x 1024, such that each row vector is a flattened 32 x 32 gray - scale image.
I'm trying to preform PCA with some K components that can be varied from 1 to the Number of features of X_centered(i, :), that is 1024. In order to get the top K eigen - vectors.
After that I'm trying to project X_centered to the K dimension using the K eigen - vectors with:
X_Projected(i, :) = [v1 v2 ... vk]^T * (X(i, :) - X_mean), i = 1, ... , 120
For some reason the "pca" Matlab function output of coeffs is 1024 x 119 and not 1024 x 1024 (The resulted eigen vectors need to be v1, ... , v1024) , is the coeffs output the eigen - vectors or I'm missing something here?