torch-geometric

You can build a graph neural network with PyTorch Geometric using prebuilt GCN, GAT, and GraphSAGE templates. The included code generator scaffolds training loops and evaluation pipelines, letting you jump straight to model development for tasks like node classification and link prediction.

Templates for graph neural networks include GCN, GAT, and GraphSAGE architectures. These prebuilt models come with integrated training loops and standard evaluation pipelines designed to accelerate prototyping for node classification, graph classification, and molecular property prediction.

Yes, PyTorch Geometric templates support both CPU and GPU execution for training graph neural networks. You need a Python environment with PyTorch and PyG installed to run the scripts and execute model training and evaluation pipelines on your hardware.

The best way to scaffold a GNN project for molecular property prediction is using a template generator. It creates ready-to-use PyG scripts with dataset and benchmark scaffolding, allowing teams to rapidly prototype models and standardize evaluation pipelines across research workflows.

Yes, the graph neural network templates require NetworkX and Matplotlib alongside PyTorch, PyG, and NumPy. These dependencies support graph data handling and visualization within the training and evaluation scripts for your node and graph classification tasks.

You can evaluate GNN models on standard benchmarks using built-in dataset scaffolding. The templates provide example usage with Planetoid and TUDataset datasets, integrating standard evaluation pipelines directly into the generated training scripts for consistent benchmarking.