GraphStorm v0.1.2 release
V0.1.2 is a minor release of GraphStorm. In this release, we add CPU support for GML model training and inference, giving users more flexibility in choosing their environment settings. We add CSV data format support for graph construction pipeline, giving users more flexibility in choosing their input data format. We add four new dataloaders for link prediction tasks to avoid the slow neighbor sampling execution path in DGL. Together with DGL 1.0.4, we can get 2.4X speedup on training a 2 layer RGCN on MAG dataset on 4 g5.48x instances. We also add two new models GLEM and Network in graph neural network in GraphStorm model zoo.
Major features
- Add CPU support for GML model training and inference. #300
- Add CSV data format support for graph construction pipeline. #324
- Speedup link prediction training. With DGL 1.0.4, we can get 2.4X speedup on training a 2 layer RGCN on MAG dataset on 4 g5.48x instances. #279 , #302
New Built-in Models
- GLEM for node-level tasks #262
- Network in graph neural network #316
Enhancements
- Optimize GraphStorm package dependencies #319
- Allow Edge classification/regression inference tasks work with graphs without test masks #298
- Add a MAE evaluation metric for regression tasks. #318
- Allow passing of SageMaker Estimator arguments during job launch #350
Contributors
- Da Zheng from AWS
- Xiang Song from AWS
- Jian Zhang from AWS
- Theodore Vasiloudis from AWS
- Prateek M Desai from AWS
- Runjie Ma from AWS
- Dominika Jedynak from Intel
- Luqi Yao from Amazon
- Israt Nisa from AWS