ogb

Why we included this project

Getting a dataset and evaluation protocol you can trust is usually the biggest hurdle in graph-ML work. OGB brings together dozens of standardized graph datasets spanning node-, link-, and graph-level prediction tasks, drawn from domains like molecular property prediction, social and information networks, and heterogeneous knowledge graphs. Its loaders plug straight into PyTorch Geometric or DGL, so a few lines of code handle the download and apply the official train/valid/test splits, and each dataset ships an evaluator that returns the standard metric. That keeps results across papers and projects directly comparable. Researchers writing graph-ML papers and engineers prototyping GNN pipelines will both find it the fastest route from a fresh idea to a credible baseline number.