K-Core based Temporal Graph Convolutional Network for Dynamic Graphs

Requirements:

• Python >= 3.6
• Numpy >= 1.17.0
• Scipy >= 1.3.0
• Pandas >= 0.25.1
• Cython >= 0.29.14
• Scikit-Learn >= 0.21.2
• Networkx >= 2.3
• Pytorch >= 1.2.0

Directory:

CTGCN/  
  CTGCN/  
    baseline/                    (implemented baselines, i.e. GCN, GAT, EvolveGCN, timers)  
    evaluation/                  (evaluation tasks, i.e. link prediction, node classification)  
    preprocessing/               (preprocessing tasks, i.e. k-core decomposition, random walk)  
    embedding.py                 (data loader and different kinds of embedding)  
    graph.py                     (dynamic graph generation for static graphs and scalability data generation)  
    layers.py                    (All layers used in CTGCN)  
    metrics.py                   (Loss function)  
    models.py                    (All models of CTGCN)  
    train.py                     (main file used to train different embedding methods)  
    utils.py                     (utility functions)  
  data/  
    facebook/  
      0. input/                  (raw data)  
      1. format/                 (formatted dynamic graph data)  
      2. embedding/              (embedding results)  
      nodes_set/                 (node list file)  
      CTGCN/                     (intermediate data, i.e. k-core data, random walk data)  
    enron/
      ......

Commands:

1. compile 'preprocessing/helper.pyx' file.

   cd preprocessing  
   python3 setup.py build_ext --inplace  
   

2. preprocess dynamic graphs to generate k-core data and random walk data.

   python3 preprocessing/__init__.py
   

3. train graph embedding methods

   python3 train.py
   

Baselines:

1. GCN, GAT, EvolveGCN, timers
   https://github.com/jhljx/CTGCN/tree/master/CTGCN/baseline
2. DeepWalk, node2vec, struc2vec
   https://github.com/shenweichen/GraphEmbedding
3. DynGEM
   https://github.com/palash1992/DynamicGEM/blob/master/dynamicgem/embedding/dynSDNE.py

1. Origin graph file names must be timestamp format or integer number format(other wise when training dynamic embedding, sorted(f_list) may reture a wrong order of files)
2. Weighted random walk are set as default in the 'get_walk_info' function of 'CTGCN/preprocessing/walk_generation.py' file.
3. The original graph edge data don't need to have a reverse edge for each edge, the procedure will add reverse edges in 'get_sp_adj_mat' / 'get_nx_graph' function of 'CTGCN/utils.py' file. All graph data sets are read by these two functions.
4. The original graph file header must be 'from_id, to_id, weight', or you will modify the 'get_nx_graph' function of 'CTGCN/utils.py' file. 'get_sp_adj_mat' don't care the concrete header name, as long as the first 2 columns are node idxs and the last column is edge weight.

• K-Core based Temporal Graph Convolutional Network for Dynamic Graphs