cheatsheet.rst

GNN Cheatsheet

• :class:`~torch_sparse.SparseTensor`: If checked (✓), supports message passing based on :class:`torch_sparse.SparseTensor`, e.g., :obj:`GCNConv(...).forward(x, adj_t)`. See here for the accompanying tutorial
• :obj:`edge_weight`: If checked (✓), supports message passing with one-dimensional edge weight information, e.g., :obj:`GraphConv(...).forward(x, edge_index, edge_weight)`
• :obj:`edge_attr`: If checked (✓), supports message passing with multi-dimensional edge feature information, e.g., :obj:`GINEConv(...).forward(x, edge_index, edge_attr)`
• bipartite: If checked (✓), supports message passing in bipartite graphs with potentially different feature dimensionalities for source and destination nodes, e.g., :obj:`SAGEConv(in_channels=(16, 32), out_channels=64)`
• static: If checked (✓), supports message passing in static graphs, e.g., :obj:`GCNConv(...).forward(x, edge_index)` with :obj:`x` having shape :obj:`[batch_size, num_nodes, in_channels]`
• lazy: If checked (✓), supports lazy initialization of message passing layers, e.g., :obj:`SAGEConv(in_channels=-1, out_channels=64)`

Graph Neural Network Operators

{% for cls in torch_geometric.nn.conv.classes[1:] %} {% if not torch_geometric.nn.conv.utils.processes_heterogeneous_graphs(cls) and

not torch_geometric.nn.conv.utils.processes_hypergraphs(cls) and not torch_geometric.nn.conv.utils.processes_point_clouds(cls) %}

• • :class:`~torch_geometric.nn.conv.{{ cls }}` (Paper)
  • {% if torch_geometric.nn.conv.utils.supports_sparse_tensor(cls) %}✓{% endif %}
  • {% if torch_geometric.nn.conv.utils.supports_edge_weights(cls) %}✓{% endif %}
  • {% if torch_geometric.nn.conv.utils.supports_edge_features(cls) %}✓{% endif %}
  • {% if torch_geometric.nn.conv.utils.supports_bipartite_graphs(cls) %}✓{% endif %}
  • {% if torch_geometric.nn.conv.utils.supports_static_graphs(cls) %}✓{% endif %}
  • {% if torch_geometric.nn.conv.utils.supports_lazy_initialization(cls) %}✓{% endif %}

{% endif %} {% endfor %}

Heterogeneous Graph Neural Network Operators

{% for cls in torch_geometric.nn.conv.classes[1:] %} {% if torch_geometric.nn.conv.utils.processes_heterogeneous_graphs(cls) %}

• • :class:`~torch_geometric.nn.conv.{{ cls }}` (Paper)
  • {% if torch_geometric.nn.conv.utils.supports_sparse_tensor(cls) %}✓{% endif %}
  • {% if torch_geometric.nn.conv.utils.supports_edge_weights(cls) %}✓{% endif %}
  • {% if torch_geometric.nn.conv.utils.supports_edge_features(cls) %}✓{% endif %}
  • {% if torch_geometric.nn.conv.utils.supports_bipartite_graphs(cls) %}✓{% endif %}
  • {% if torch_geometric.nn.conv.utils.supports_static_graphs(cls) %}✓{% endif %}
  • {% if torch_geometric.nn.conv.utils.supports_lazy_initialization(cls) %}✓{% endif %}

{% endif %} {% endfor %}

Hypergraph Neural Network Operators

{% for cls in torch_geometric.nn.conv.classes[1:] %} {% if torch_geometric.nn.conv.utils.processes_hypergraphs(cls) %}

• • :class:`~torch_geometric.nn.conv.{{ cls }}` (Paper)
  • {% if torch_geometric.nn.conv.utils.supports_sparse_tensor(cls) %}✓{% endif %}
  • {% if torch_geometric.nn.conv.utils.supports_edge_weights(cls) %}✓{% endif %}
  • {% if torch_geometric.nn.conv.utils.supports_edge_features(cls) %}✓{% endif %}
  • {% if torch_geometric.nn.conv.utils.supports_bipartite_graphs(cls) %}✓{% endif %}
  • {% if torch_geometric.nn.conv.utils.supports_static_graphs(cls) %}✓{% endif %}
  • {% if torch_geometric.nn.conv.utils.supports_lazy_initialization(cls) %}✓{% endif %}

{% endif %} {% endfor %}

Point Cloud Neural Network Operators

{% for cls in torch_geometric.nn.conv.classes[1:] %} {% if torch_geometric.nn.conv.utils.processes_point_clouds(cls) %}

• • :class:`~torch_geometric.nn.conv.{{ cls }}` (Paper)
  • {% if torch_geometric.nn.conv.utils.supports_bipartite_graphs(cls) %}✓{% endif %}
  • {% if torch_geometric.nn.conv.utils.supports_lazy_initialization(cls) %}✓{% endif %}

{% endif %} {% endfor %}