Closed DaniJonesOcean closed 2 years ago
Now that I have my code working on BAS HPC, I have access to more memory and can try to train Kernel PCA.
So far, I've been able to train Kernel PCA using about 1/3 of the entire dataset. Here's the pairplot for a 3D kernel PCA:
It still looks kind of "spiky". For comparison, here's the standard PCA:
We tried it. No obvious advantages. Closing now.
It is a bit 'spiky' perhaps... Have you tried using some sort of kernel PCA?
_Originally posted by @maikejulie in https://github.com/so-wise/weddell_gyre_clusters/issues/6#issuecomment-919569100_