search

ndey96 / GCNN-Explainability

Implementation of "Explainability Methods for Graph Convolutional Neural Networks" from HRL Laboratories
MIT License
75 stars 3 forks source link

GCN explainability for Cora/Pubmed/citeseer #3

Open abhigoku10 opened 3 years ago

abhigoku10 commented 3 years ago

@ndey96 thanks for sharing the code base had one query , the current code base implementation is for chem dataset can we apply this to citeseer / pubmed / cora dataset and check its results ?? if so what all changes have to be done do give some references? thanks in adavance

ndey96 commented 3 years ago

Hi @abhigoku10 if you would like to try out some other datasets you need to implement a load_citeseer function similar to the load_bbbp function in data.py. You need to preprocess the dataset to extract node features (X), adjacency matrix (A), and labels. The dataset should be a list of torch_geometric.data.Data objects

abhigoku10 commented 3 years ago

@ndey96 thanks for the response , i tried to implement load_citeseer similar to load_bbbp but failing in many cases since the current code is tightly coupled with bbbp is there any other references which i can use for node classification dataset

ndey96 commented 3 years ago

@abhigoku10 can you provide more information? How is the code tightly coupled to bbbp? How did you fail specifically?