SAGEConvStack

class openff.nagl.nn.gcn.SAGEConvStack(*args, _is_dgl: bool = False, **kwargs)[source]

Bases: BaseGCNStack[Union[SAGEConv, dgl.nn.pytorch.SAGEConv]]

GraphSAGE graph convolutional neural network for atom embeddings.

GraphSAGE GCNs learn a function that iteratively improves a node embedding by mixing in aggregated feature vectors of progressively more distant neighborhoods. GraphSAGE is inductive, scales to large graphs, and makes good use of feature-rich node embeddings.

Layers in this network use the DGL SAGEConv class.

See also

dgl.nn.pytorch.conv.SAGEConv

Attributes

available_aggregator_types

default_activation_function

default_aggregator_type

default_dropout

name

training

available_aggregator_types = ['mean', 'gcn', 'pool', 'lstm']
default_activation_function = <class 'torch.nn.modules.activation.ReLU'>
default_aggregator_type = 'mean'
default_dropout = 0.0
name = 'SAGEConv'
training: bool