Installation¶

This page takes you from a clean environment to a working GraphNetz install. It assumes familiarity with PyTorch and PyG; everything else is covered as we go.

Install¶

uv add graphnetz
# or, in an existing environment:
pip install graphnetz

Requires Python ≥ 3.10, PyTorch ≥ 2.6, and torch-geometric ≥ 2.6.

Optional extras¶

Extra	Install	Unlocks
`ogb`	`pip install graphnetz[ogb]`	OGB loaders (`ogbn-arxiv`, `ogbl-collab`, `ogbn-products`, `ogbg-molhiv`, `ogbg-molpcba`)
`chem`	`pip install graphnetz[chem]`	RDKit — required by OGB molecular loaders such as `ogbg-molhiv`

Development install¶

For local development, clone the repo and use the dev group:

git clone https://github.com/Kleyt0n/graphnetz
cd graphnetz
uv sync --group dev

Verify¶

from graphnetz import GCN, train_node_classification
from graphnetz.datasets.social import cora

ds = cora("data/cora")
model = GCN(ds.num_features, 64, ds.num_classes)
history = train_node_classification(model, ds[0], epochs=10)

Tip

GPU is automatic. Both the standalone trainers and run_benchmark accept device='auto' (the default). The runtime picks CUDA when available, then Apple-silicon MPS, then CPU, and moves the model and data onto it for you. Pin placement explicitly with device='cpu' (or any torch.device) when you need to.

Next: the Quickstart runs your first multi-seed benchmark.