correct usage

[ekg]

I'm observing that the structures I would expect in PC1 and PC2 are ending up in PC2 and PC3.

I suspect it's a problem with how I'm using the system.

Do you have several examples where you compare the output of PCA with your method to alternative implementations? I'd like to work back from a case that you're sure is set up right.

[ekg]

Here is what I'm looking at:

Left is Randomized SVD as implemented in this library.

Right is PCA from R on the same data.

I can share code to do this, but I'm thinking there is something so obvious that I'm worried it is just a bug in my code and I don't want you to chase it down if I can just work from validated example.

[ekg]

And for good measure, here is PC1 vs PC2 from randomized SVD.