You prepare the data, and CogDL does everything else.
A Unified Trainer
CogDL integrates a unified trainer with decoupled modules for the GNN training. Based on this unique design, CogDL can provide extensive features such as hyperparameter optimization, distributed training, training techniques, and experiment management.
Efficient Sparse Operators
Efficiency is one of the most significant characteristics of CogDL. CogDL develops well-optimized sparse kernel operators to speed up the training of GNN models, enabling it become the most competitive graph libraries for efficiency.
Ease of Use
We provide simple APIs in CogDL such that users only need to write one line of code to train and evaluate any graph representation learning methods. In addition, CogDL also collects and maintains the state-of-the-art configurations, facilitating open, robust, and reproducible deep learning research on graphs.