Build, train, and debug graph neural networks with PyTorch Geometric.
npx skills add https://github.com/shushuzn/Rairos --skill torch-geometric-shushuzn
Please help me install this Agent Skill. Skill: torch-geometric Source: https://github.com/shushuzn/Rairos/tree/main/skills/torch-geometric Command: npx skills add https://github.com/shushuzn/Rairos --skill torch-geometric-shushuzn
It removes the guesswork in implementing Graph Neural Networks by providing the right PyTorch Geometric (PyG) data structures, layer patterns, and training workflows for graph learning tasks.
Data and HeteroData so you can represent homogeneous and heterogeneous graphs correctly, including edge_index handling.MessagePassing, and scaling training with NeighborLoader / LinkNeighborLoader.to_hetero, HeteroConv, and HGTConv.torch_geometric.explain with algorithms like GNNExplainer and CaptumExplainer for model interpretation.Use the skill when you need to implement or debug a PyTorch Geometric GNN for tasks like node classification or link prediction from your graph tensors (including correct edge_index construction).
High-intent search queries and answers about installing and using this skill.
To build a GNN for node classification, use PyTorch Geometric data structures like `Data` for graph tensors, stack common convolution layers, and apply training patterns designed for node-level tasks. Ensure correct `edge_index` construction to represent graph topology accurately.
Handle heterogeneous graphs in PyG using `HeteroData` structures and build type-aware models with `to_hetero`, `HeteroConv`, or `HGTConv`. This allows distinct node and edge types to be processed correctly within your graph neural network architecture.
The `edge_index` defines graph connectivity for message passing layers and must be a tensor of shape `[2, num_edges]` specifying source and target nodes. Correct `edge_index` semantics are crucial for proper information aggregation across graph edges.
Scale GNN training on large graphs using `NeighborLoader` for node-level tasks and `LinkNeighborLoader` for link prediction. These PyG loaders enable scalable mini-batch workflows by sampling localized subgraphs instead of loading the entire graph into memory.
Yes, you can implement custom message passing logic by subclassing the `MessagePassing` base class. This allows you to define custom aggregation and update functions for creating specialized graph convolution layers tailored to your specific graph learning task.
Explain GNN predictions using the `torch_geometric.explain` module with algorithms like `GNNExplainer` and `CaptumExplainer`. These tools provide model interpretability by identifying important nodes and edges that influence your graph neural network's outputs.
tondevrelFacilitate deep learning on graph-structured data with PyTorch Geometric.
hxk622Develop and train Graph Neural Networks using the PyTorch Geometric library.
hung-phanBuild and train PyTorch deep learning models with custom training loops.
Zeyuyang-0420Build and optimize Graph Neural Networks with PyTorch Geometric.