Summary

In version 3.1.0, Angel enhances graph learning ability, and adds a variety of improvements, including:

• Features in graph learning with the trend of graph data structure adopted for many applications such as social network analysis and recommendation systems
• Publishing a collection of well implemented graph algorithms such as traditional learning, graph embedding, and graph deep learning - These algorithms can be used directly in the production model by calling with simple configurations
• Providing an operator API for graph manipulations including building graph, and operating the vertices and edges
• Enabling the support of GPU devices within the PyTorch-on-Angel running mode - With this feature it’s possible to leverage the hardwares to speed up the computation intensive algorithms

New Features

• Traditional Graph Learning Algorithms

  • TriangleCountUndirected is used to determine the number of triangles passing through each node in a undirected graph.
  • CC calculate Connected Components over graph.
  • LPA can detect communities in networks through label propagation process.
  • PageRank is generally used for node importance evaluation.
  • K-core algorithm is used to separate and extract closely connected subgraphs in the graph.
  • H-index algorithm calculates the h-index value for each node in a undirected graph, it's usually used to represent the importance of a vertex.
  • Closeness The Closeness Centrality of a node measures its average farness(inverse diastance) to all other nodes.This algorithm aim to detecting nodes that are able to spread information very efficiently through a graph.
  • CommonFriends is used to calculate the common friend number of two users，and can be used to measure the closeness and used in recommendation and security system.
  • Louvain algorithm is a classic community discovery algorithm, which optimizes the modularity index to achieve the purpose of community division.

• Graph Embedding

  • LINE is a network representation learning algorithm, which can also be considered as a preprocessing algorithm for graph data.
  • Word2Vec can learn the vector representation of words from text data or node form graph data.

• Graph Deep Learning

  • GraphSage learns vertex embedding by training the function that aggregates the neighbors' info of the vertex, which plays a general role for unknown nodes.
  • R-GCN are related to a recent class of neural networks operating on graphs, and are developed specifically to deal with the highly multi-relational data characteristic of realistic knowledge bases.
  • DGI is a general approach for learning node representations within graph-structured data in an unsupervised manner.

• Basic Graph Operators

  • Graph Bulider
  • Graph Information
  • Sampling Operator

• GPU Support in PyTorch-on-Angel

  • Enable GPU in PyTorch on Angel to accelerated graph deep learning algorithm.