OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Evaluation & Observability

Open Graph Benchmark

Open Graph Benchmark (OGB) is an open-source Python library offering benchmark datasets, data loaders and standardized evaluators for graph machine learning. It supports node-, link- and graph-level prediction tasks and documents loaders for PyTorch Geometric and DGL.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Open Graph Benchmark — OpenViST project profile in AI Evaluation & Observability.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use OGB when a graph-learning experiment needs benchmark data, standardized splits and a dataset-specific evaluator, particularly when working with PyTorch Geometric or DGL.

Who it is for

  • Researchers comparing graph-machine-learning methods using common datasets and evaluation protocols.
  • Developers building graph-learning experiments in PyTorch Geometric or DGL.
  • Teams preparing reproducible benchmark splits for node, link or graph prediction.

Documented uses

  • Download and prepare benchmark graph datasets with standardized splits.
  • Create training, validation and test data loaders for graph-learning experiments.
  • Evaluate model predictions with metrics defined for a benchmark dataset.
  • Compare methods across graph prediction tasks and dataset domains.
Before you choose

Requirements & limits

OGB is identified as MIT-licensed open-source software. The documented installation method is pip.

  • The documented framework options are PyTorch Geometric and DGL; the requirements list either framework alongside PyTorch.
  • The README requires Python 3.6 or later and PyTorch 1.6 or later, plus NumPy, pandas, urllib3, scikit-learn and outdated.
  • Compute needs depend on dataset scale: small datasets may fit on one GPU, while medium and large graphs may require multiple GPUs or sampling/partitioning.
  • Evaluators expect a dataset-specific input format and return the metric appropriate to that dataset.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Prepare graph-property benchmark data with PyTorch Geometric

Download the documented ogbg-molhiv dataset and create loaders using its standardized training, validation and test splits.

Before you start

  • Python 3.6 or later
  • PyTorch 1.6 or later
  • PyTorch Geometric 2.0.2 or later
  • OGB installation via pip; the README also lists NumPy, pandas, urllib3, scikit-learn and outdated as requirements.

Steps

  1. Install OGB with Python’s package manager: pip install ogb.

  2. Import PygGraphPropPredDataset from ogb.graphproppred and DataLoader from torch_geometric.loader.

  3. Create PygGraphPropPredDataset with name='ogbg-molhiv'; the example downloads and processes the dataset at ./dataset/ogbg_molhiv/.

  4. Get the split indices with dataset.get_idx_split(), then create separate DataLoader instances for the train, valid and test indices.

Expected outcome

A downloaded and processed graph-property dataset with loaders for the standardized training, validation and test splits.

Guide limitations

  • This recipe covers dataset preparation only; it does not define or train a model.
  • The README notes that medium- and large-scale graphs may need multiple GPUs or sampling/partitioning techniques.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
15 October 2026
Original website title
snap-stanford/ogb · Official homepage title · retained in its source language
How we review sourcesSuggest a correction