​

https://github.com/benedekrozemberczki/walklets

Walklets is a multi-scale node embedding algorithm which learns an embedding of approximated adjacency matrix powers up to a given order. Walklet places nodes in an abstract feature space where the vertex features are able to reproduce connectivity patterns in the graph at multiple scales. Embedding is created with exponential implicit factorization. Feature vectors that are extracted in an unsupervised way can be used in downstream machine learning tasks such as edge prediction, node classification and community detection.

​

https://preview.redd.it/z75hl07du3w11.jpg?width=1438&format=pjpg&auto=webp&v=enabled&s=e80a2582812ecbad6039e4e79dc4aefa49613126

The implementation supports second-order random walk sampling, which was proposed in the original paper but was not implemented in it. The second-order random walks sampling methods were taken from the reference implementation of https://github.com/aditya-grover/node2vec.