Closed larsoner closed 3 years ago
@larsoner cool! Once it is implemented we should try to understand its benefits :)
Once it is implemented we should try to understand its benefits :)
It's implemented here, feel free to try it out on some data
This method is indeed valid and has been used in multiple application papers.
This implements symmetric orthogonalization following https://www.sciencedirect.com/science/article/pii/S1053811915002670.
@agramfort @dengemann @SherazKhan any thoughts on the utility of this versus the already-implemented pairwise orthogonalization? It seems to have the limitation of requiring n_labels < rank, but some advantages mentioned in the paper above (better sensitivity / fewer false alarms among them).