The StellarGraph team is happy to announce the release of StellarGraph version 0.10.
StellarGraph is an open-source library featuring state-of-the-art graph machine learning algorithms. The project is delivered as part of CSIRO’s Data61.
Dramatically improved memory usage is the key feature of the 0.10 release, with the StellarGraph and StellarDiGraph classes now backed by NumPy and Pandas leading to significant performance benefits.
- Link prediction with directed GraphSAGE
- GraphWave, which computes structural node embeddings by using wavelet transforms on the graph Laplacian.
Other new algorithms and features remain under active development, but are available in this release as experimental previews:
- Temporal Random Walks: random walks that respect the time that each edge occurred (stored as edge weights)
- Watch Your Step: computes node embeddings by simulating the effect of random walks, rather than doing them.
- ComplEx: computes embeddings for nodes and edge types in knowledge graphs, and uses these to perform link prediction
- Neo4j connector: the GraphSAGE algorithm can execute neighbourhood sampling from a Neo4j database, so the edges of a graph do not have to fit into memory.
Major improvements and fixes:
- StellarGraph now supports TensorFlow 2.1
- Demos now focus on Jupyter notebooks
- Supervised GraphSAGE Node Attribute Inference algorithm is now reproducible
- Code for saliency maps/interpretability refactored to have more sharing, making it cleaner and easier to extend
- Demo notebooks predominantly tested on CI using Papermill, so won’t go out of date.
See the full release notes here.