mgplvm.rdist.GPbase module

class mgplvm.rdist.GPbase.GPbase(manif, m, n_samples, ts, _scale=0.9, ell=None)[source]

Bases: mgplvm.rdist.common.Rdist

I_v(v, sample_idxs=None)[source]: Compute I @ v for some vector v. This should be implemented for each class separately. v is (n_samples x d x m x n_mc) where n_samples is the number of sample_idxs

K_half(sample_idxs=None)[source]: compute one column of the square root of the prior matrix

concentration_parameters()[source]

property ell: torch.Tensor

Return type: Tensor

full_cov()[source]: Compute the full covariance Khalf @ I @ I @ Khalf

gmu_parameters()[source]

kl(batch_idxs=None, sample_idxs=None)[source]: Compute KL divergence between prior and posterior. This should be implemented for each class separately

property lat_mu: return variational mean mu = K_half @ nu

msg(Y=None, batch_idxs=None, sample_idxs=None)[source]

name = 'GPbase'

property nu: torch.Tensor

Return type: Tensor

property prms

sample(size, Y=None, batch_idxs=None, sample_idxs=None, kmax=5, analytic_kl=False, prior=None)[source]: generate samples and computes its log entropy

property scale: torch.Tensor

Return type: Tensor

training: bool