gammagl.models.AGNNModel

class AGNNModel(feature_dim, hidden_dim, num_class, n_att_layers, dropout_rate, is_cora=False, name=None)[source]

The graph attention operator from the “Attention-based Graph Neural Network for Semi-supervised Learning” paper.

Parameters:

  • feature_dim (int) – Dimension of feature vector in original input.

  • hidden_dim (int) – Dimension of feature vector in AGNN.

  • num_class (int) – Dimension of feature vector in forward output.

  • n_att_layers (int) – Number of attention layers.

  • dropout_rate (float) – Dropout rate.

  • is_cora (bool, optional) – Whether the dateset is cora. There is a special operation on cora that cora dataset contains two agnn_conv layers.

  • name (str, optional) – The name of the model.

forward(x, edge_index, num_nodes)[source]