PyTorch Geometric (PyG) — Graph Neural Networks

PyTorch Geometric (PyG) — Graph Neural Networks

Overview

PyTorch Geometric is a library built on PyTorch for developing and training Graph Neural Networks (GNNs). It provides 40+ convolutional layers, mini-batch processing via block-diagonal adjacency matrices, neighbor sampling for large-scale graphs, and heterogeneous graph support for multi-type node/edge networks.

When to Use

• Node classification on citation, social, or biological networks
• Graph-level classification (molecular activity, protein function)
• Link prediction (knowledge graphs, recommendation systems)
• Molecular property prediction (drug discovery, quantum chemistry)
• 3D point cloud processing and mesh analysis
• Large-scale graph learning with neighbor sampling (>100K nodes)
• Heterogeneous graphs with multiple node/edge types
• For non-graph deep learning → use PyTorch directly
• For traditional graph algorithms (shortest path, centrality) → use NetworkX

Prerequisites

pip install torch torch_geometric
# Optional sparse operations (recommended):
# pip install pyg_lib torch_scatter torch_sparse torch_cluster

import torch
import torch.nn.functional as F
from torch_geometric.data import Data
from torch_geometric.nn import GCNConv

Quick Start

from torch_geometric.datasets import Planetoid
from torch_geometric.nn import GCNConv
import torch, torch.nn.functional as F