Welcome to DeepGNN’s documentation!

DeepGNN is a package for training/evaluating ML models on graph data. It is a Python library that provides:

• A graph engine object designed for ML tasks with an assortment of routines for sampling nodes, edges and neighbors as well as feature fetching.
• Various aggregators, encoders and decoders to pass graph data to neural nets.
• Basic NN layers for training: convolution, attention and bindings to pytorch-geometric library.
• A collection of trainers to work with models in local and distributed environments.

Documentation

• Graph Engine
  • Graph engine overview
  • Preparing Graph Data
  • How to create graph engine from a networkx graph
  • How to Create a Custom Decoder from Avro to Binary
  • Distributed conversion
  • Temporal Graph Support in DeepGNN
  • Edge List Format
  • Binary conversion
  • Python API
• Pytorch
  • GraphSage Link Prediction Model
  • Node Classification with GAT
  • Distributed training of a GraphSage model
• Tensorflow
  • Node Classification with GAT
  • Custom Link Prediction Model
  • Ray Usage Example for Node Classification with GAT
• Advanced
  • HDFS Streaming
  • Example of using SQL database to work with graphs.