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

Bases: Module

Model for Graph Classification task

Parameters:

  • dataset – ([Graph]): List of input graphs

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

  • conv – (str): Name of the convolution used for Neural Network

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

  • dropout – (int): Dropout (default: 0)

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

  • ssl_flag – (bool): If True, self supervised loss would be alsooptimized during the training, in addition to semi-supervised

  • heads – (int): Number of heads in GAT layer

static convert_dataset(data: List[Graph], train_indices: List[int], val_indices: List[int]) → Tuple[List[Graph], List[Graph], List[Graph], int]

Convert input dataset to train,test, val according to provided indices

Parameters:

  • data – ([Graph]): List of graphs as input dataset

  • train_indices – ([int]): List of indices for train dataset

  • val_indices – ([int]): List of indices for validation dataset

Returns:

([Graph],[Graph],[Graph], int): Lists of train and validation graphs and the minimum size among all graphs

forward(x: torch_geometric.typing.Tensor, edge_index: torch_geometric.typing.Adj, batch: torch_geometric.typing.Tensor) → Tuple[torch_geometric.typing.Tensor, torch_geometric.typing.Tensor]

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

Parameters:

  • x – (Tensor) Input features

  • edge_index – (Adj) Edge index of a batch

  • batch – Batch of data

Returns:

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

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

Negative log likelihood loss

Parameters:

  • pred – (Tensor): Predicted labels

  • label – (Tensor): Genuine labels

Returns:

(Tensor): Loss

static self_supervised_loss(deg_pred: torch_geometric.typing.Tensor, batch: torch_geometric.data.Batch) → torch_geometric.typing.Tensor

Self Supervised Loss for Graph Classsification task, MSE between predicted average degree of each graph and genuine ones

Parameters:

  • deg_pred – (Tensor): Tensor of predicted degrees of graphs in dataset

  • batch – (): Batch of train data

Returns:

(Tensor): Loss