The code for our ICLR paper: StructPool: Structured Graph Pooling via Conditional Random Fields
=============
The code is built based on DGCNN(https://github.com/muhanzhang/pytorch_DGCNN) and Graph UNet(https://github.com/HongyangGao/Graph-U-Nets). Thanks a lot for their code sharing!
We first employ GCNs to obtain u(x) for a batch. Next, perform pooling for each graph.
Please refer to "DGCNN_embedding.py" and "pool.py" for details.
This implementation is based on Hanjun Dai's structure2vec graph backend. Under the "lib/" directory, type
make -j4
to compile the necessary c++ files.
After that, under the root directory of this repository, type
./run_DGCNN.sh
to run DGCNN on dataset MUTAG with the default setting.
Or type
./run_DGCNN.sh DATANAME FOLD
to run on dataset = DATANAME using fold number = FOLD (1-10, corresponds to which fold to use as test data in the cross-validation experiments).
If you set FOLD = 0, e.g., typing "./run_DGCNN.sh DD 0", then it will run 10-fold cross validation on DD and report the average accuracy.
Alternatively, type
./run_DGCNN.sh DATANAME 1 200
to use the last 200 graphs in the dataset as testing graphs. The fold number 1 will be ignored.
Check "run_DGCNN.sh" for more options.
The first step is to transform your graphs to the format described in "data/README.md". You should put your testing graphs at the end of the file. Then, there is an option -test_number X, which enables using the last X graphs from the file as testing graphs. You may also pass X as the third argument to "run_DGCNN.sh" by
./run_DGCNN.sh DATANAME 1 X
where the fold number 1 will be ignored.
@inproceedings{
Yuan2020StructPool:,
title={StructPool: Structured Graph Pooling via Conditional Random Fields},
author={Hao Yuan and Shuiwang Ji},
booktitle={International Conference on Learning Representations},
year={2020},
url={https://openreview.net/forum?id=BJxg_hVtwH}
}