DrugEncoder.GAT

Click here to view source code.

class GAT(x_num_embedding: int = 178, edge_num_embedding: int = 18, embedding_dim: int = 768,
ft_dim: int = 768, num_heads: int = 4, dropout: float = 0.3, hid_dim: int = 384, num_layers: int = 2)

GAT can be used to encode graphs of drugs.

PARAMETERS:

  • x_num_embedding (int, optional) - Number of node embeddings. (default: 178)

  • edge_num_embedding (int, optional) - Number of edge embeddings. (default: 18)

  • embedding_dim (int, optional) - Dimension of embeddings. (default: 768)

  • ft_dim (int, optional) - Dimension of encoded node features. (default: 768)

  • num_heads (int, optional) - Parameter heads of torch_geometric.nn.conv.GATConv. (default: 4)

  • dropout (float, optional) - Dropout rate of torch_geometric.nn.conv.GATConv. (default: 0.3)

  • hid_dim (int, optional) - Dimension of hidden layers. (default: 384)

  • num_layers (int, optional) - Number of torch_geometric.nn.conv.GATConv layers. (default: 3)

SHAPES:

  • input: Preprocessed graphs

  • output: (Encoded node features [node_num, ft_dim], Preprocessed graphs)

forward(g)

  • g - The input of GAT.