pelican.layers.GeneralAggregator

class pelican.layers.GeneralAggregator(aggregator, in_rank, out_rank, in_channels, out_channels, map_multipliers=True, factorize=False, aggr='mean')

Bases: Module

General aggregator class.

Parameters:

• aggregator (callable) – Aggregation function to use.

• in_rank (int) – Input rank (0 for graph, 1 for nodes, 2 for edges).

• out_rank (int) – Output rank (0 for graph, 1 for nodes, 2 for edges).

• in_channels (int) – Number of input channels.

• out_channels (int) – Number of output channels.

• map_multipliers (bool) – Whether to use learnable multipliers for each aggregation map, by default True.

• factorize (bool) – Whether to use factorized coefficients, by default False. Factorization reduces the number of parameters.

• aggr (str) – Aggregation method to use (‘mean’, ‘sum’, ‘prod’, ‘amin’, ‘amax’), by default ‘mean’.

property coeffs

forward(x, *args, **kwargs)

Forward pass of the aggregator.

Parameters:

• x (torch.Tensor) – Input features of shape (in_objects, in_channels).

• *args – Additional arguments to pass to the aggregator.

• **kwargs – Additional keyword arguments to pass to the aggregator.

Returns:

out – Output features of shape (out_objects, in_channels).

Return type:

torch.Tensor