torch-geometric
Build Graph Neural Networks with PyTorch Geometric
Also available from: davila7
Graph Neural Networks power modern applications like drug discovery, social network analysis, and recommendation systems. This skill provides ready-to-use templates and documentation for implementing GNNs with PyTorch Geometric.
Download the skill ZIP
Upload in Claude
Go to Settings → Capabilities → Skills → Upload skill
Toggle on and start using
Test it
Using "torch-geometric". Create a GCN model for the Cora citation network
Expected outcome:
- GCNConv layers with hidden channels
- Training loop with loss calculation
- Evaluation on train/val/test masks
- Model saved to checkpoint file
Using "torch-geometric". Load the TUDataset for graph classification
Expected outcome:
- TUDataset with ENZYMES loaded to /tmp
- DataLoader with batch_size=32
- Global mean pooling layer
- Training across 100 epochs
Using "torch-geometric". Use attention mechanisms in my GNN
Expected outcome:
- GATConv with 8 attention heads
- Dropout applied to attention coefficients
- ELU activation function
- Multi-head output concatenated then averaged
Security Audit
SafeAll 777 static findings are false positives. The detected 'eval()' calls are legitimate PyTorch model.eval() methods for evaluation mode, not dynamic code execution. The 'external_commands' findings are markdown code blocks using backtick formatting. The 'weak cryptographic algorithm' flags are triggered by documentation mentioning algorithm names. The 'certificate/key files' flags misidentify error messages referencing TEMPLATES.keys(). No actual security threats identified in this legitimate PyTorch Geometric ML skill.
Risk Factors
⚙️ External commands (4)
📁 Filesystem access (2)
⚡ Contains scripts (3)
🌐 Network access (1)
Quality Score
What You Can Build
Build Molecular Property Predictors
Train GNNs on molecular graphs to predict drug properties, toxicity, and binding affinity for drug discovery.
Create Citation Network Classifiers
Implement node classification on citation networks like Cora and PubMed for paper categorization and recommendation.
Analyze Social Networks
Apply graph attention and sampling techniques to large social networks for community detection and influence analysis.
Try These Prompts
Build a Graph Convolutional Network for node classification on the Cora dataset using PyTorch Geometric. Include training loop and evaluation.
Create a GraphSAGE model for graph classification on the ENZYMES dataset using global mean pooling and DataLoader.
Implement a Graph Attention Network (GAT) with multi-head attention for citation network classification.
Show how to use NeighborLoader for training GNNs on large-scale graphs with mini-batch processing and neighbor sampling.
Best Practices
- Use model.eval() and torch.no_grad() during inference to disable gradients and ensure deterministic behavior.
- Apply NeighborLoader with appropriate num_neighbors for large graphs to enable efficient mini-batch training.
- Save models using torch.save() with state_dict for efficient checkpointing and reproducibility.
Avoid
- Avoid training on full graphs without sampling for large networks as this causes memory overflow.
- Do not forget to set model.train() during training phases and model.eval() during evaluation.
- Avoid hardcoding dataset paths; use root parameter with temporary directories for dataset caching.
Frequently Asked Questions
What is PyTorch Geometric?
What GNN architectures are supported?
What datasets can I load?
How do I handle large graphs?
Can I do molecular property prediction?
What transforms are available?
Developer Details
Author
K-Dense-AILicense
MIT license
Repository
https://github.com/K-Dense-AI/claude-scientific-skills/tree/main/scientific-skills/torch_geometricRef
main
File structure