grb.model.dgl

grb.model.dgl.gat

class grb.model.dgl.gat.GAT(in_features, out_features, hidden_features, n_layers, n_heads, activation=<function leaky_relu>, layer_norm=False, feat_norm=None, adj_norm_func=None, feat_dropout=0.0, attn_dropout=0.0, residual=False, dropout=0.0)[source]

Bases: Module

Graph Attention Networks (GAT)

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.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: None.

  • feat_dropout (float, optional) – Dropout rate for input features. Default: 0.0.

  • attn_dropout (float, optional) – Dropout rate for attention. Default: 0.0.

  • 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

property model_name
property model_type
training: bool

grb.model.dgl.gcn

class grb.model.dgl.gcn.GCN(in_features, out_features, hidden_features, activation=<function relu>, layer_norm=False)[source]

Bases: Module

forward(x, adj, dropout=0)[source]

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

property model_type
training: bool

grb.model.dgl.gin

class grb.model.dgl.gin.ApplyNodeFunc(mlp)[source]

Bases: Module

Update the node feature hv with MLP, BN and ReLU.

forward(h)[source]

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool
class grb.model.dgl.gin.GIN(in_features, hidden_features, out_features, learn_eps=True, neighbor_pooling_type='sum', num_mlp_layers=1)[source]

Bases: Module

GIN model

forward(x, adj, dropout=0)[source]

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

property model_type
training: bool
class grb.model.dgl.gin.MLP(num_layers, input_dim, hidden_dim, output_dim)[source]

Bases: Module

MLP with linear output

forward(x)[source]

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

grb.model.dgl.grand

class grb.model.dgl.grand.GRAND(in_features, out_features, hidden_features, n_layers=2, s=1, k=3, temp=1.0, lam=1.0, feat_norm=None, adj_norm_func=None, node_dropout=0.0, input_dropout=0.0, hidden_dropout=0.0)[source]

Bases: Module

Graph Random Neural Networks (GRAND)

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.

  • s (int) – Number of Augmentation samples

  • k (int) – Number of Propagation Steps

  • node_dropout (float) – Dropout rate on node features.

  • input_dropout (float) – Dropout rate of the input layer of a MLP

  • hidden_dropout (float) – Dropout rate of the hidden layer of a MLP

forward(x, adj)[source]

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

property model_name
property model_type
training: bool
grb.model.dgl.grand.GRANDConv(graph, features, hop)[source]
Parameters

  • graph (dgl.Graph) – The input graph

  • features (torch.Tensor) – Tensor of node features

  • hop (int) – Propagation Steps

grb.model.dgl.grand.drop_node(features, drop_rate, training=True)[source]