scglue.models.sc.GraphDecoder.forward

GraphDecoder.forward(v, eidx, esgn)[source]

Decode graph from vertex latent

Parameters:

  • v (torch.Tensor) – Vertex latent (\(n_{vertices} \times n_{features}\))

  • eidx (torch.Tensor) – Vertex indices of edges (\(2 \times n_{edges}\))

  • esgn (torch.Tensor) – Sign of edges (\(n_{edges}\))

Return type:

torch.distributions.bernoulli.Bernoulli

Returns:

recon – Edge reconstruction distribution (\(n_{edges}\))