grb.model.torch¶

grb.model.torch.appnp¶

Torch module for APPNP.

class grb.model.torch.appnp.APPNP(in_features, out_features, hidden_features, n_layers, layer_norm=False, activation=<function relu>, edge_drop=0.0, alpha=0.01, k=10, feat_norm=None, adj_norm_func=<function GCNAdjNorm>, dropout=0.0)[source]¶

Bases: Module

Approximated Personalized Propagation of Neural Predictions (APPNP)

Parameters

in_features (int) – Dimension of input features.
out_features (int) – Dimension of output features.
hidden_features (int or list of int) – Dimension of hidden features. List if multi-layer.
n_layers (int) – Number of layers.
layer_norm (bool, optional) – Whether to use layer normalization. Default: False.
activation (func of torch.nn.functional, optional) – Activation function. Default: torch.nn.functional.relu.
feat_norm (str, optional) – Type of features normalization, choose from [“arctan”, “tanh”, None]. Default: None.
adj_norm_func (func of utils.normalize, optional) – Function that normalizes adjacency matrix. Default: GCNAdjNorm.
edge_drop (float, optional) – Rate of edge drop.
alpha (float, optional) – Hyper-parameter, refer to original paper. Default: 0.01.
k (int, optional) – Hyper-parameter, refer to original paper. Default: 10.
dropout (float, optional) – Dropout rate during training. Default: 0.0.

forward(x, adj)[source]¶

Parameters

x (torch.Tensor) – Tensor of input features.
adj (torch.SparseTensor) – Sparse tensor of adjacency matrix.

Returns

x – Output of model (logits without activation).

Return type

torch.Tensor

property model_name¶

property model_type¶: Indicate type of implementation.

reset_parameters()[source]¶: Reset parameters.

training: bool¶

class grb.model.torch.appnp.SparseEdgeDrop(edge_drop)[source]¶

Bases: Module

Sparse implementation of edge drop.

Parameters: edge_drop (float) – Rate of edge drop.

forward(adj)[source]¶: Sparse edge drop

training: bool¶

grb.model.torch.gcn¶

Torch module for GCN.

class grb.model.torch.gcn.GCN(in_features, out_features, hidden_features, n_layers, activation=<function relu>, layer_norm=False, residual=False, feat_norm=None, adj_norm_func=<function GCNAdjNorm>, dropout=0.0)[source]¶

Bases: Module

Graph Convolutional Networks (GCN)

Parameters

in_features (int) – Dimension of input features.
out_features (int) – Dimension of output features.
hidden_features (int or list of int) – Dimension of hidden features. List if multi-layer.
n_layers (int) – Number of layers.
layer_norm (bool, optional) – Whether to use layer normalization. Default: False.
activation (func of torch.nn.functional, optional) – Activation function. Default: torch.nn.functional.relu.
residual (bool, optional) – Whether to use residual connection. Default: False.
feat_norm (str, optional) – Type of features normalization, choose from [“arctan”, “tanh”, None]. Default: None.
adj_norm_func (func of utils.normalize, optional) – Function that normalizes adjacency matrix. Default: GCNAdjNorm.
dropout (float, optional) – Dropout rate during training. Default: 0.0.

forward(x, adj)[source]¶

Parameters

x (torch.Tensor) – Tensor of input features.
adj (torch.SparseTensor) – Sparse tensor of adjacency matrix.

Returns

x – Output of model (logits without activation).

Return type

torch.Tensor

property model_name¶

property model_type¶: Indicate type of implementation.

reset_parameters()[source]¶: Reset parameters.

training: bool¶

class grb.model.torch.gcn.GCNConv(in_features, out_features, activation=None, residual=False, dropout=0.0)[source]¶

Bases: Module

GCN convolutional layer.

Parameters

in_features (int) – Dimension of input features.
out_features (int) – Dimension of output features.
activation (func of torch.nn.functional, optional) – Activation function. Default: None.
residual (bool, optional) – Whether to use residual connection. Default: False.
dropout (float, optional) – Dropout rate during training. Default: 0.0.

forward(x, adj)[source]¶

Parameters

x (torch.Tensor) – Tensor of input features.
adj (torch.SparseTensor) – Sparse tensor of adjacency matrix.

Returns

x – Output of layer.

Return type

torch.Tensor

reset_parameters()[source]¶: Reset parameters.

training: bool¶

class grb.model.torch.gcn.GCNGC(in_features, out_features, hidden_features, n_layers, activation=<function relu>, layer_norm=False, residual=False, feat_norm=None, adj_norm_func=<function GCNAdjNorm>, dropout=0.0)[source]¶

Bases: Module

Graph Convolutional Networks (GCN)

Parameters

in_features (int) – Dimension of input features.
out_features (int) – Dimension of output features.
hidden_features (int or list of int) – Dimension of hidden features. List if multi-layer.
n_layers (int) – Number of layers.
layer_norm (bool, optional) – Whether to use layer normalization. Default: False.
activation (func of torch.nn.functional, optional) – Activation function. Default: torch.nn.functional.relu.
residual (bool, optional) – Whether to use residual connection. Default: False.
feat_norm (str, optional) – Type of features normalization, choose from [“arctan”, “tanh”, None]. Default: None.
adj_norm_func (func of utils.normalize, optional) – Function that normalizes adjacency matrix. Default: GCNAdjNorm.
dropout (float, optional) – Dropout rate during training. Default: 0.0.

forward(x, adj, batch_index=None)[source]¶

Parameters

x (torch.Tensor) – Tensor of input features.
adj (torch.SparseTensor) – Sparse tensor of adjacency matrix.

Returns

x – Output of model (logits without activation).

Return type

torch.Tensor

property model_name¶

property model_type¶: Indicate type of implementation.

reset_parameters()[source]¶: Reset parameters.

training: bool¶

grb.model.torch.gin¶

Torch module for GIN.

class grb.model.torch.gin.GIN(in_features, out_features, hidden_features, n_layers, n_mlp_layers=2, activation=<function relu>, layer_norm=False, batch_norm=True, eps=0.0, feat_norm=None, adj_norm_func=None, dropout=0.0)[source]¶

Bases: Module

Graph Isomorphism Network (GIN)

Parameters

in_features (int) – Dimension of input features.
out_features (int) – Dimension of output features.
hidden_features (int or list of int) – Dimension of hidden features. List if multi-layer.
n_layers (int) – Number of layers.
n_mlp_layers (int) – Number of layers.
layer_norm (bool, optional) – Whether to use layer normalization. Default: False.
batch_norm (bool, optional) – Whether to apply batch normalization. Default: True.
eps (float, optional) – Hyper-parameter, refer to original paper. Default: 0.0.
activation (func of torch.nn.functional, optional) – Activation function. Default: torch.nn.functional.relu.
feat_norm (str, optional) – Type of features normalization, choose from [“arctan”, “tanh”, None]. Default: None.
adj_norm_func (func of utils.normalize, optional) – Function that normalizes adjacency matrix. Default: None.
dropout (float, optional) – Rate of dropout. Default: 0.0.

forward(x, adj)[source]¶

Parameters

x (torch.Tensor) – Tensor of input features.
adj (torch.SparseTensor) – Sparse tensor of adjacency matrix.

Returns

x – Output of model (logits without activation).

Return type

torch.Tensor

property model_name¶

property model_type¶: Indicate type of implementation.

reset_parameters()[source]¶: Reset parameters.

training: bool¶

class grb.model.torch.gin.GINConv(in_features, out_features, activation=<function relu>, eps=0.0, batch_norm=True, dropout=0.0)[source]¶

Bases: Module

GIN convolutional layer.

Parameters

in_features (int) – Dimension of input features.
out_features (int) – Dimension of output features.
activation (func of torch.nn.functional, optional) – Activation function. Default: None.
eps (float, optional) – Hyper-parameter, refer to original paper. Default: 0.0.
batch_norm (bool, optional) – Whether to apply batch normalization. Default: True.
dropout (float, optional) – Rate of dropout. Default: 0.0.

forward(x, adj)[source]¶

Parameters

x (torch.Tensor) – Tensor of input features.
adj (torch.SparseTensor) – Sparse tensor of adjacency matrix.

Returns

x – Output of layer.

Return type

torch.Tensor

reset_parameters()[source]¶: Reset parameters.

training: bool¶

grb.model.torch.graphsage¶

Torch module for GraphSAGE.

class grb.model.torch.graphsage.GraphSAGE(in_features, out_features, hidden_features, n_layers, activation=<function relu>, layer_norm=False, feat_norm=None, adj_norm_func=<function SAGEAdjNorm>, mu=2.0, dropout=0.0)[source]¶

Bases: Module

Inductive Representation Learning on Large Graphs (GraphSAGE)

Parameters

in_features (int) – Dimension of input features.
out_features (int) – Dimension of output features.
n_layers (int) – Number of layers.
hidden_features (int or list of int) – Dimension of hidden features. List if multi-layer.
layer_norm (bool, optional) – Whether to use layer normalization. Default: False.
activation (func of torch.nn.functional, optional) – Activation function. Default: torch.nn.functional.relu.
feat_norm (str, optional) – Type of features normalization, choose from [“arctan”, “tanh”, None]. Default: None.
adj_norm_func (func of utils.normalize, optional) – Function that normalizes adjacency matrix. Default: SAGEAdjNorm.
mu (float, optional) – Hyper-parameter, refer to original paper. Default: 2.0.
dropout (float, optional) – Rate of dropout. Default: 0.0.

forward(x, adj)[source]¶

Parameters

x (torch.Tensor) – Tensor of input features.
adj (torch.SparseTensor) – Sparse tensor of adjacency matrix.

Returns

x – Output of model (logits without activation).

Return type

torch.Tensor

property model_name¶

property model_type¶: Indicate type of implementation.

reset_parameters()[source]¶: Reset parameters.

training: bool¶

class grb.model.torch.graphsage.SAGEConv(in_features, pool_features, out_features, activation=None, mu=2.0, dropout=0.0)[source]¶

Bases: Module

SAGE convolutional layer.

Parameters

in_features (int) – Dimension of input features.
pool_features (int) – Dimension of pooling features.
out_features (int) – Dimension of output features.
activation (func of torch.nn.functional, optional) – Activation function. Default: None.
dropout (float, optional) – Rate of dropout. Default: 0.0.
mu (float, optional) – Hyper-parameter, refer to original paper. Default: 2.0.

forward(x, adj)[source]¶

Parameters

x (torch.Tensor) – Tensor of input features.
adj (torch.SparseTensor) – Sparse tensor of adjacency matrix.

Returns

x – Output of layer.

Return type

torch.Tensor

reset_parameters()[source]¶: Reset parameters.

training: bool¶

grb.model.torch.robustgcn¶

grb.model.torch.sgcn¶

Torch module for SGCN.

class grb.model.torch.sgcn.SGCN(in_features, out_features, hidden_features, n_layers, activation=<built-in method tanh of type object>, feat_norm=None, adj_norm_func=<function GCNAdjNorm>, layer_norm=False, batch_norm=False, k=4, dropout=0.0)[source]¶

Bases: Module

Simplifying Graph Convolutional Networks (SGCN)

Parameters

in_features (int) – Dimension of input features.
out_features (int) – Dimension of output features.
hidden_features (int or list of int) – Dimension of hidden features. List if multi-layer.
n_layers (int) – Number of layers.
layer_norm (bool, optional) – Whether to use layer normalization. Default: False.
activation (func of torch.nn.functional, optional) – Activation function. Default: torch.tanh.
k (int, optional) – Hyper-parameter, refer to original paper. Default: 4.
feat_norm (str, optional) – Type of features normalization, choose from [“arctan”, “tanh”, None]. Default: None.
adj_norm_func (func of utils.normalize, optional) – Function that normalizes adjacency matrix. Default: GCNAdjNorm.
dropout (float, optional) – Rate of dropout. Default: 0.0.

forward(x, adj)[source]¶

Parameters

x (torch.Tensor) – Tensor of input features.
adj (torch.SparseTensor) – Sparse tensor of adjacency matrix.

Returns

x – Output of model (logits without activation).

Return type

torch.Tensor

property model_name¶

property model_type¶: Indicate type of implementation.

training: bool¶

class grb.model.torch.sgcn.SGConv(in_features, out_features, k)[source]¶

Bases: Module

SGCN convolutional layer.

Parameters

in_features (int) – Dimension of input features.
out_features (int) – Dimension of output features.
k (int, optional) – Hyper-parameter, refer to original paper. Default: 4.

Returns

x – Output of layer.

Return type

torch.Tensor

forward(x, adj)[source]¶

Parameters

x (torch.Tensor) – Tensor of input features.
adj (torch.SparseTensor) – Sparse tensor of adjacency matrix.

Returns

x – Output of layer.

Return type

torch.Tensor

training: bool¶

grb.model.torch.tagcn¶

Torch module for TAGCN.

class grb.model.torch.tagcn.TAGCN(in_features, out_features, hidden_features, n_layers, k, activation=<function leaky_relu>, feat_norm=None, adj_norm_func=<function GCNAdjNorm>, layer_norm=False, batch_norm=False, dropout=0.0)[source]¶

Bases: Module

Topological Adaptive Graph Convolutional Networks (TAGCN)

Parameters

in_features (int) – Dimension of input features.
out_features (int) – Dimension of output features.
hidden_features (int or list of int) – Dimension of hidden features. List if multi-layer.
n_layers (int) – Number of layers.
k (int) – Hyper-parameter, k-hop adjacency matrix, refer to original paper.
layer_norm (bool, optional) – Whether to use layer normalization. Default: False.
batch_norm (bool, optional) – Whether to apply batch normalization. Default: False.
activation (func of torch.nn.functional, optional) – Activation function. Default: torch.nn.functional.leaky_relu.
feat_norm (str, optional) – Type of features normalization, choose from [“arctan”, “tanh”, None]. Default: None.
adj_norm_func (func of utils.normalize, optional) – Function that normalizes adjacency matrix. Default: GCNAdjNorm.
dropout (float, optional) – Rate of dropout. Default: 0.0.

forward(x, adj)[source]¶

Parameters

x (torch.Tensor) – Tensor of input features.
adj (torch.SparseTensor) – Sparse tensor of adjacency matrix.

Returns

x – Output of model (logits without activation).

Return type

torch.Tensor

property model_name¶

property model_type¶: Indicate type of implementation.

reset_parameters()[source]¶: Reset paramters.

training: bool¶

class grb.model.torch.tagcn.TAGConv(in_features, out_features, k=2, activation=None, batch_norm=False, dropout=0.0)[source]¶

Bases: Module

TAGCN convolutional layer.

Parameters

in_features (int) – Dimension of input features.
out_features (int) – Dimension of output features.
k (int, optional) – Hyper-parameter, refer to original paper. Default: 2.
activation (func of torch.nn.functional, optional) – Activation function. Default: None.
batch_norm (bool, optional) – Whether to apply batch normalization. Default: False.
dropout (float, optional) – Rate of dropout. Default: 0.0.

forward(x, adj)[source]¶

Parameters

x (torch.Tensor) – Tensor of input features.
adj (torch.SparseTensor) – Sparse tensor of adjacency matrix.

Returns

x – Output of layer.

Return type

torch.Tensor

reset_parameters()[source]¶: Reset parameters.

training: bool¶