class NegativeSampler(data: Graph, device: torch.device, loss_info: Dict)

Bases: BaseSamplerWithNegative

Sampler for positive and negative edges using random walk based methods

Parameters:

• data – (Graph): Input dataset

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

• loss_info – (dict): Dict of parameters of unsupervised loss function

negative_sample(batch: torch_geometric.data.Batch) → Tuple[torch_geometric.typing.Tensor, torch_geometric.typing.Tensor]

Sample positive and negative edges for batch nodes

Parameters:

batch – (Batch): Nodes for positive and negative sampling from them

Returns:

(Tensor, Tensor): positive and negative samples

class SamplerAPP(data: Graph, device: torch.device, loss_info: Dict)

Bases: BaseSamplerWithNegative

Sample positive and negative edges for APP unsupervised loss function

Parameters:

• data – (Graph): Input dataset

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

• loss_info – (dict): Dict of parameters of unsupervised loss function

sample(batch: torch_geometric.data.Batch) → Tuple[torch_geometric.typing.Tensor, torch_geometric.typing.Tensor]

Sample positive and negative edges for batch of nodes

Parameters:

batch – (Batch): Batch for which sampling should be conducted

Returns:

(Tensor,Tensor): positive and negative edges

class SamplerContextMatrix(data: Graph, device: torch.device, loss_info: Dict)

Bases: BaseSamplerWithNegative

Sample positive and negative edges for context matrix based unsupervised loss function

Parameters:

• data – (Graph): Input dataset

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

• loss_info – (dict): Dict of parameters of unsupervised loss function

class SamplerFactorization(data: Graph, device: torch.device, loss_info: Dict)

Bases: BaseSampler

Sample positive and negative edges for context matrix based unsupervised loss function

Parameters:

• data – (Graph): Input dataset

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

• loss_info – (dict): Dict of parameters of unsupervised loss function

sample(batch: torch_geometric.typing.Tensor) → torch_geometric.typing.Tensor

Sample of positive and negative edges for Graph Factorization-based unsupervosed loss functions

Parameters:

batch – (Batch): Nodes for which sampling should be conducted

Returns:

(Tensor): Positive and negative edges

class SamplerRandomWalk(data: Graph, device: torch.device, loss_info: Dict)

Bases: BaseSamplerWithNegative

Sampler for positive and negative edges using random walk based methods

Parameters:

• data – (Graph): Input dataset

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

• loss_info – (dict): Dict of parameters of unsupervised loss function