Filtration-Enhanced Graph Transformation

Dependencies

• python (==3.8)
• torch (==1.10.0)
• pyg (==2.0.2)
• torch-scatter (==1.5.9)
• torch-sparse (==0.6.12)
• networkx (==2.6.2)
• GraphRicciCurvature (==0.5.3)
• igraph (==0.9.8)
• tabulate (==0.8.9)
• GraKeL (==0.1.8)
• numpy (==1.21.0)
• pandas (==1.2.5)
• sklearn (==0.24.2)
• scipy (==1.7.0)

Data

Add dataset name in tmp_ds.txt which is available in https://ls11-www.cs.tu-dortmund.de/staff/morris/graphkerneldataset. And we have provided build_all_feg.sh and build_all_fes.shin sh to download all data and build filtration-enhanced graphs and snapshots. To download dataset and build FEG with filtration of native edge weight(or native vertex attributes), navigate to sh folder and type the following command into the terminal:

$ ./build_all_feg.sh attr
$ ./build_all_feg.sh vattr

Run Graph Kernels for FEG

To run Weisfeiler-Leman Subtree kernel, ShortestPath Kernel and GraphLet Kernel for datasets with filtration of native edge weight, navigate to sh folder and type the following command into the terminal:

$ ./run_kernel.sh attr [directory name]

Run Graph Kernels for FES

To run Weisfeiler-Leman Subtree kernel, ShortestPath Kernel and GraphLet Kernel for datasets with filtration of native vertex attributes, navigate to sh folder and type the following command into the terminal:

$ ./run_kernel_snapshot.sh vattr [directory name]

Run GIN

To run GIN for datasets with filtration of core number, navigate to sh folder and type the following command into the terminal:

$ ./run_gin.sh gin [lr] [dropout] degeneracy

Run GraphSAGE

To run GIN for datasets with filtration of ricci-curvature, navigate to sh folder and type the following command into the terminal:

$ ./run_sage.sh [lr] [dropout] curvature

Run GraphSNN

To run GraphSNN for datasets with filtration of ricci-curvature, navigate to sh folder and type the following command into the terminal:

$ ./run_graphsnn.sh [lr] [dropout] curvature