Code for experiment in Derivation of Back-propagation for Graph Convolutional Networks using Matrix Calculus and its Application to Explainable Artificial Intelligence
For Figure 5, the code is located at Code/Node classification/5-layer GCN/(2-class) (loop) GCN_KarateClub_Kronecker_compare.py
For Figure 7, the code is located at Code/Link prediction/5-layer GCN/(small_general_loop) Kronecker_vs_auto_5_layer.py To run (small_general_loop) Kronecker_vs_auto_5_layer.py, the user needs to download the DDI_100_nodes.pkl located at Code/Link prediction/ and replace the string on the right-hand side of the equal sign at the 10th line of the code to the file location of your downloaded DDI_100_nodes.pkl.
For Figure 1 (a), Figure 2, Figure 3 (a), and Figure 4 (a), the code is located at Code/Node classification/1-layer GCN/2 class of GCN_KarateClub_feature.ipynb
For Figure 1 (b), Figure 3 (b), Figure 4 (b), Figure 6, the code is located at Code/Node classification/2-layer GCN/Small_sens_of_Link_prediction_Kronecker_vs_auto_drug.ipynb The file DDI_100_nodes.pkl located at Code/Link prediction/ needs to be uploaded when runing Small_sens_of_Link_prediction_Kronecker_vs_auto_drug.ipynb.