spamosaic.architectures.gat.GATConv

class spamosaic.architectures.gat.GATConv(*args: Any, **kwargs: Any)

Graph Attention Network (GAT) layer adapted from STAGATE implementation.

Parameters:

• in_channels (int or tuple of int) – Dimension(s) of input node features.

• out_channels (int) – Dimension of output node features.

• heads (int, default=1) – Number of attention heads.

• concat (bool, default=True) – Whether to concatenate multi-head outputs (True) or average them (False).

• negative_slope (float, default=0.2) – LeakyReLU angle of the negative slope.

• dropout (float, default=0.0) – Dropout probability for attention coefficients.

• add_self_loops (bool, default=True) – Whether to add self-loops to the input graph.

• bias (bool, default=True) – Whether to add bias (not used here).

• **kwargs (optional) – Additional arguments for MessagePassing.