Code for PermGNN
This directory contains code necessary for running PermGNN experiments.
For more information, please check our paper:
Roy, Indradyumna, Abir De, and Soumen Chakrabarti. "Adversarial Permutation Guided Node Representations for Link Prediction." arXiv preprint arXiv:2012.08974 (2020). [PDF]
Data used in the paper is hosted at [DATA]
Recent versions of Pytorch, numpy, scipy, sklearn, networkx and matplotlib are required. You can install all the required packages using the following command:
$ pip install -r requirements.txt
To train PermGNN on the datasets- Twitter/Google/PB/cora/citeseer
python -m minmax.linkpred_main --TASK="PermGNN" --want_cuda=True --BATCH_SIZE=128 --TEST_FRAC=0.2 --VAL_FRAC=0.1 --DIR_PATH="." --LEARNING_RATE_FUNC=0.0005 --LEARNING_RATE_PERM=0.0005 --MARGIN=0.001 --OPTIM="Adam" --SCORE="MAP" --DATASET_NAME="Twitter_3"
python -m minmax.linkpred_main --TASK="PermGNN" --want_cuda=True --BATCH_SIZE=128 --TEST_FRAC=0.2 --VAL_FRAC=0.1 --DIR_PATH="." --LEARNING_RATE_FUNC=0.001 --LEARNING_RATE_PERM=0.001 --MARGIN=0.01 --OPTIM="Adam" --SCORE="MAP" --DATASET_NAME="Gplus_1"
python -m minmax.linkpred_main --TASK="PermGNN" --want_cuda=True --BATCH_SIZE=128 --TEST_FRAC=0.2 --VAL_FRAC=0.1 --DIR_PATH="." --LEARNING_RATE_FUNC=0.001 --LEARNING_RATE_PERM=0.001 --MARGIN=0.01 --OPTIM="Adam" --SCORE="MAP" --DATASET_NAME="PB"
python -m minmax.linkpred_main --TASK="PermGNN" --want_cuda=True --BATCH_SIZE=128 --TEST_FRAC=0.2 --VAL_FRAC=0.1 --DIR_PATH="." --LEARNING_RATE_FUNC=0.001 --LEARNING_RATE_PERM=0.001 --MARGIN=0.1 --OPTIM="Adam" --SCORE="MAP" --DATASET_NAME="citeseer"
python -m minmax.linkpred_main --TASK="PermGNN" --want_cuda=True --BATCH_SIZE=128 --TEST_FRAC=0.2 --VAL_FRAC=0.1 --DIR_PATH="." --LEARNING_RATE_FUNC=0.001 --LEARNING_RATE_PERM=0.001 --MARGIN=0.01 --OPTIM="Adam" --SCORE="MAP" --DATASET_NAME="cora"To run and compare performnce of our proposed hashing technique with random hyperplane hashing and exhaustive enumeration
python -m minmax.hashing_main --DATASET_NAME="citeseer"
python -m minmax.hashing_main --DATASET_NAME="cora"
python -m minmax.hashing_main --DATASET_NAME="Twitter_3"
python -m minmax.hashing_main --DATASET_NAME="Gplus_1"
python -m minmax.hashing_main --DATASET_NAME="PB"To reproduct the plots from paper for training loss variation wrt Ktau and Input/Output sensitivity
python -m minmax.performance_variation_wrt_ktau --TEST_FRAC=0.2 --VAL_FRAC=0.1 --DATASET_NAME="cora"
python -m minmax.performance_variation_wrt_ktau --TEST_FRAC=0.2 --VAL_FRAC=0.1 --DATASET_NAME="citeseer"- GPU usage is required
- source code files are all in minmax folder.
- Paste the contents [DATA] in data folder and run the code.
If you find the code useful, please cite our paper:
@article{roy2020adversarial,
title={Adversarial Permutation Guided Node Representations for Link Prediction},
author={Roy, Indradyumna and De, Abir and Chakrabarti, Soumen},
journal={arXiv preprint arXiv:2012.08974},
year={2020}
}
Indradyumna Roy, Indian Institute of Technology - Bombay
indraroy15@cse.iitb.ac.in