Bering.models.EdgeClf.forward

EdgeClf.forward(z_node, data, num_pos_edges, num_neg_edges, image, conv2d_padding=10)[source]

Run the decoder model from latent space z. Before running the decoder, random positive and negative edges are generated as the input.

Parameters:

  • z – Latent features from pretrained node classification (n samples x n latent features)

  • data (Tensor) – Input data loader (several graphs)

  • num_pos_edges (int) – Number of positive edges

  • num_neg_edges (int) – Number of negative edges

  • image (Tensor) – Image tensor for computing the conv2d embedding

  • conv2d_padding (int) – add paddings in the conv2d embedding