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

Bases: Module

Model for Node Classification task with Layer, considering grph characteristics

Parameters:

  • dataset – (Graph): Input Graph

  • device – (device): Device ‘cuda’ or ‘cpu’

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

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

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

  • ssl_flag – (bool): If True, self supervised loss will be optimized additionally to semi-supervised (default: False)

  • loss_name – (str): Name of loss function for embedding learning in GeomGCN layer

inference(data: Graph) → Tuple[torch_geometric.typing.Tensor, torch_geometric.typing.Tensor]

Count the representation of the node on the next layer of the model

Parameters:

data – (Graph): Input Graph

Returns:

(Tensor, Tensor): Predicted probabilities of labels and predicted degrees of nodes

static loss_sup(pred: torch_geometric.typing.Tensor, label: torch_geometric.typing.Tensor) → torch_geometric.typing.Tensor

Count negative log likelihood loss function

Parameters:

  • pred – (Tensor): Predicted labels

  • label – (Tensor): Genuine labels

Returns:

(Tensor): Loss

self_supervised_loss(deg_pred: torch_geometric.typing.Tensor, dat: torch_geometric.typing.Tensor) → torch_geometric.typing.Tensor

Self supervised loss, predicting degrees of nodes

Parameters:

  • deg_pred – (Tensor): Predicted degrees

  • dat – (Tensor): Train mask

Returns:

(Tensor): Loss