Data exploration examines and summarizes node features and graph structure before modeling. It helps reveal topology, relationships, and potential blockers to effective GNN training. section 4.2.2 section 4.2.3
A practical graph exploration workflow includes:
- Load labels, node features, and an adjacency list from the MATLAB dataset. section 4.2.2
- Convert the adjacency list into a NetworkX graph and inspect node count, edge count, and average degree. section 4.2.3
- Analyze connected components, density, clustering coefficients, and degree distributions to understand topology and modeling constraints. section 4.2.3
Listing 4.1 Load data
prefix = 'PATH_TO_MATLAB_FILE/' data_file = loadmat(prefix + 'YelpChi.mat') #1 labels = data_file['label'].flatten() #2 features = data_file['features'].todense().A #2 yelp_homo = data_file['homo'] #3 sparse_to_adjlist(yelp_homo, prefix +\ 'yelp_homo_adjlists.pickle')
Use the loading listing to reproduce the initial dataset preparation, including MATLAB import, feature extraction, and adjacency-list serialization. section 4.2.2
YelpChi graph profile
Exploration finds 45,954 nodes, 3,892,933 edges, an average degree of 84.71, 26 connected components, sparse density, and an average clustering coefficient of 0.77. The largest component contains nearly the entire graph. section 4.2.3
