Changelog • torchgnn

torchgnn 0.1.0

First release.
Message-passing layers: layer_gcn(), layer_gcn_general(), layer_sage(), layer_gat(), layer_gin() and layer_regconv().
Multi-layer models: model_gcn(), model_gcn_general(), model_sage(), model_gat() and model_gin().
Graph-level pooling with pool_global_add(), pool_global_mean() and pool_global_max(), and normalization with layer_layer_norm().
Graph construction and preparation helpers: adj_from_edgelist(), nodes_to_tensor(), gcn_normalize(), adj_row_normalize(), add_graph_self_loops() and graph_split().
Requires ‘torch’ 0.17 or later, which removes the need for manual 1-based index adjustments. Related https://github.com/mlverse/torch/issues/1460.