ForeCA

[N1r]

I found out this pretty repo for foreca dyca and cca, I wonder why foreca has been limited to d < 1 , for what kind of reason? thanks.

def init(self, d, T, init="random_ortho", n_init=1, tol=1e-6,
verbose=False, device="cpu", dtype=torch.float64,
rng_or_seed=20200818):
self.d = d
if d > 1:
raise ValueError

[JesseLivezey]

Hi @N1r, we didn't have the need to implement this for d > 1, so we never got around to it. Foreca for d > 1 is a little more tricky than other methods because you need to keep the projection orthonormal during optimization. This is easy for d = 1, but requires more work for d > 1.