Can I create custom message-passing layers?

Absolutely. The skill explains how to subclass the MessagePassing class and utilize the propagate, message, and aggregate functions.

What is PyTorch Geometric (PyG)?

PyG is a library built on PyTorch designed for deep learning on graphs and other irregular structures, providing a wide range of GNN layers and data handling utilities.

Does this skill support heterogeneous graphs?

Yes, it includes specialized guidance for HeteroData objects and layers like HGTConv and RGCNConv designed for multiple node and edge types.

How do I handle large graphs that don't fit in GPU memory?

This skill provides implementation patterns for neighbor sampling and mini-batch loaders like NeighborLoader, which sample subgraphs to enable scalable training.

PyTorch Geometric (PyG) GNN Builder

Name: PyTorch Geometric (PyG) GNN Builder
Author: K-Dense-AI

byK-Dense-AI

•

17,327

•

データサイエンスとML

Builds, trains, and scales Graph Neural Networks using the PyTorch Geometric (PyG) library for complex relational data.

This skill provides expert guidance for implementing Graph Neural Networks (GNNs) using PyTorch Geometric, the leading library for geometric deep learning. It assists developers and researchers in defining graph data structures, configuring message-passing layers, and implementing advanced patterns like link prediction, heterogeneous graph learning, and large-scale neighbor sampling. Whether you are performing node classification, graph-level analysis, or molecular modeling, this skill ensures your implementation follows PyG best practices for performance and scalability.

主な機能

01Automated dataset handling and preprocessing transforms

02Implementation patterns for GCN, GAT, GraphSAGE, and GIN architectures

0317,327 GitHub stars

04Comprehensive guidance on Data and HeteroData structures

05Custom GNN layer development via the MessagePassing base class

06Scalable training solutions using NeighborLoader for massive datasets

ユースケース

01Building molecular property prediction models for scientific research

02Scaling GNN training on large social or citation networks using mini-batching

03Developing recommendation systems using heterogeneous graph link prediction

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add k-dense-ai/claude-scientific-skills torch-geometric

For use in Claude.ai and ChatGPT

主な機能

01Automated dataset handling and preprocessing transforms

02Implementation patterns for GCN, GAT, GraphSAGE, and GIN architectures

0317,327 GitHub stars

04Comprehensive guidance on Data and HeteroData structures

05Custom GNN layer development via the MessagePassing base class

06Scalable training solutions using NeighborLoader for massive datasets

ユースケース

01Building molecular property prediction models for scientific research

02Scaling GNN training on large social or citation networks using mini-batching

03Developing recommendation systems using heterogeneous graph link prediction

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add k-dense-ai/claude-scientific-skills torch-geometric

For use in Claude.ai and ChatGPT