HMPNN

HMPNN class.

class topomodelx.nn.hypergraph.hmpnn.HMPNN(in_channels, hidden_channels, n_layers=2, adjacency_dropout_rate=0.7, regular_dropout_rate=0.5, **kwargs)

Neural network implementation of HMPNN [1].

Parameters:

in_channelsint

Dimension of input features.

hidden_channelsTuple[int]

A tuple of hidden feature dimensions to gradually reduce node/hyperedge representations feature dimension from in_features to the last item in the tuple.

n_layersint, default = 2

Number of HMPNNLayer layers.

adjacency_dropout_rateint, default = 0.7

Adjacency dropout rate.

regular_dropout_rateint, default = 0.5

Regular dropout rate applied on features.

**kwargsoptional

Additional arguments for the inner layers.

References

[1]

Heydari S, Livi L. Message passing neural networks for hypergraphs. ICANN 2022. https://arxiv.org/abs/2203.16995

forward(x_0, x_1, incidence_1)

Forward computation through layers.

Parameters:

x_0torch.Tensor, shape = (n_nodes, in_features)

Node features.

x_1torch.Tensor, shape = (n_hyperedges, in_features)

Hyperedge features.

incidence_1torch.sparse.Tensor, shape = (n_nodes, n_hyperedges)

Incidence matrix (B1).

Returns:

x_0torch.Tensor

Output node features.

x_1torch.Tensor

Output hyperedge features.