divelab / DIG

A library for graph deep learning research
https://diveintographs.readthedocs.io/
GNU General Public License v3.0
1.84k stars 281 forks source link

A question about DeepLift #106

Closed for-just-we closed 2 years ago

for-just-we commented 2 years ago

Does the implemented DeepLift provide edge-level explanation result for each edge? I wonder whether DeepLift could be reimplemented in a node-level version, that calculation a score for each node.

for-just-we commented 2 years ago

Another question, could xxgraph support dgl, the PyG version I use only support homogeneous graph, dgl seems to support Heterogeneous graph.

CM-BF commented 2 years ago

Thank you for your questions.

The vanilla implementation of DeepLIFT is at the node level. And the node score is mapped to edge scores naively. If you want to use node scores, I suggest you install our repo locally by pip install -e . and modified the output. Happy hacking.

The answer to the second question is yes. We are going to release a new version DIG that will support the latest PyG version. However, we don't have a plan to support dgl, currently.

for-just-we commented 2 years ago

Thank you for your questions.

The vanilla implementation of DeepLIFT is at the node level. And the node score is mapped to edge scores naively. If you want to use node scores, I suggest you install our repo locally by pip install -e . and modified the output. Happy hacking.

The answer to the second question is yes. We are going to release a new version DIG that will support the latest PyG version. However, we don't have a plan to support dgl, currently.

Thank you, I have one more question. could xxgraph support heterogeneous graph?