divide into random subspaces

i have an image slice of size 16*150. After applying SIFT algorithm i got descriptors in 128 dimensions. on that SIFT feature vector i have done PCA(principal component analysis) and i got eigen vectors and eigen values for that feature vector. Now i have to divide them into 10 random subspaces. i dont know how to do that.. please anybody help me.. In the refernce paper it has given that,for each slice, perform PCA on the training set and then keep all the eigenvectors with non-zero eigenvalues as candidates to construct 10 random PCA subspaces , each spanned by 300 PCA dimensions. The first 200 dimensions are fixed according to the first 200 eigenvectors with the largest eigenvalues, which preserve most of the facial variation. The other 100 dimensions are randomly selected from the remaining eigenvectors, which are used to capture the local facial details. please anybody explain me about this..

[nagini]

please anyone help me with this step..