class BaseNet(*args: Any, **kwargs: Any)

Bases: Module, ABC

The model for learning latent embeddings in unsupervised manner for Geom-GCN layer

Parameters:
  • device – (device): Either ‘cuda’ or ‘cpu’

  • hidden_layer – (int): The size of hidden layer (default:64)

  • out_layer – (int): The size of output layer (default:128)

  • dropout – (float): Dropout (default:0.0)

  • num_layers – (int): Number of layers in the model (default:2)

  • heads – (int): Number of heads in GAT conv (default:1)

forward(x: torch_geometric.typing.Tensor, adjs: torch_geometric.loader.neighbor_sampler.EdgeIndex) torch_geometric.typing.Tensor

Find representations of the node

Parameters:
  • x – (Tensor): Features of nodes

  • adjs – (EdgeIndex): Edge indices of computational graph for each layer

Returns:

(Tensor): Representations of nodes

inference(data: Graph, dp: float = 0) torch_geometric.typing.Tensor

Count representations of the node

Parameters:
  • data – (Graph): Input data

  • dp – (float): Dropout (default:0.0)

Returns:

(Tensor): Representations of nodes

abstract loss(out: torch_geometric.typing.Tensor, pos_neg_samples: torch_geometric.typing.Tensor) torch_geometric.typing.Tensor

Calculate loss

Parameters:
  • out – Tensor

  • pos_neg_samples – Tensor

Returns:

(Tensor) Loss

reset_parameters() None

Reset parameters