PCA First Test

Just completed the first run of tests on my implementation of PCA. Testing it agains the MATLAB implementation and it seems to get pretty similar results:

The tests were performed on a 2000 data point swiss roll data set with original dimensionality of 3.  The reduced data set had dimensionality 2.  Firstly the MATLAB version….

Secondly my own…..

The location of individual data samples seems to be different in places but the overall shape of the reduced data is correct.  My initial thoughts are that this is to do with the fact that the eigendecomposition method I use returns the second eigenvector in a negative direction.  This means that when it is multiplied with the eigenvalues and eventually the zero mean data it results in different data points to that of the MATLAB data.