🎱 tgp is out!

Created on July 25, 2025

2025

We released Torch Geometric Pool (🎱 tgp), the library for pooling in Graph Neural Networks!

About TGP

Torch Geometric Pool is a comprehensive library built on top of PyTorch Geometric that brings every major graph-pooling operator into a single, unified framework. With just a few lines of code, you can turn any vanilla GNN into a hierarchical one using drop-in pooling layers.

Key Features

The main features include:

  • 📦 Unified API for all pooling operators following the SRC (Select, Reduce, Connect) framework
  • 🧮 18 pooling methods available right now, with more to come including:
    • Sparse techniques like Top-K, NDPPooling, GraclusPooling
    • Dense methods such as Diffpool and MinCutPool
  • ⚡️ Significant speed-up thanks to precomputed graph coarsening capabilities
  • 🧩 Seamless integration with GNNs built in PyTorch Geometric
  • Alias-based instantiation for quick setup (e.g., "topk", "ndp", "diffpool", "mincut")

Resources