torch-geometric

To build a Graph Neural Network for node classification, use PyTorch Geometric's Data structures to load datasets like Cora, then construct a model such as a two-layer GCN to train, evaluate, and predict node labels.

The correct edge_index format for graph data structures in PyTorch Geometric is a tensor of shape [2, num_edges] defining source and target nodes. Correct formatting is essential to avoid common pitfalls during GNN execution.

Yes, PyTorch Geometric supports heterogeneous graphs using the HeteroData container. It handles distinct node and edge types, provides auto-conversion tools, and enables GNN training across complex relational schemas.

To scale Graph Neural Network training for large graphs, apply PyTorch Geometric's neighbor sampling strategies and scalable loaders. This enables mini-batch training on massive graphs without loading the entire topology into memory.

Yes, you can perform link prediction and graph classification end-to-end with PyTorch Geometric. It provides specialized GNN layers, transforms, and training utilities to handle these specific graph machine learning tasks.

Common pitfalls when designing GNNs with PyTorch Geometric include incorrect edge_index formatting, improper handling of heterogeneous graph relations, and overlooking scalability constraints. The Skill provides guidance to enforce correct API patterns and explainability.