-
Notifications
You must be signed in to change notification settings - Fork 26
Description
In #411 one of the main pieces of data is a block diagonal matrix of eigenvectors.
When we sort by eigenvalue and truncate to some number of top components I believe this matrix loses its structure.
Because it was only one matrix I cheated and got around this by expanding it to a dense representation. This works but is somewhat unsatisfying.
One way to manage this might be to use masks and index translation, but that should be wrapped up in a class. Essentially population selection would not really change the underlying matrix, rather defining the mask/swapping.
I believe this area of the legacy MATLAB code was somewhat bespoke, while we are trying to maintain some more uniformity in ASPIRE-Python. I am open to other ways of managing this...