mashaan14/RPTree-GCN

Random Projection Forest Initialization for Graph Convolutional Networks

arXiv preprint

@misc{https://doi.org/10.48550/arxiv.2302.12001,
	url 	  = {https://arxiv.org/abs/2302.12001},  	
  	title 	  = {Random Projection Forest Initialization for Graph Convolutional Networks},
	author 	  = {Alshammari, Mashaan and Stavrakakis, John and Ahmed, Adel F. and Takatsuka, Masahiro},
  	publisher = {arXiv},
  	year 	  = {2023}
}

paper

@article{ALSHAMMARI2023102315,
	title 	= {Random Projection Forest Initialization for Graph Convolutional Networks},
	author 	= {Mashaan Alshammari and John Stavrakakis and Adel F. Ahmed and Masahiro Takatsuka}
	journal = {MethodsX},
	year 	= {2023},
	doi 	= {https://doi.org/10.1016/j.mex.2023.102315},	
}

Files that we modified from the original GCN code

• \GCN\RPTree.py
  • a code that returns an rpTree given a feature matrix X
• \GCN\utils.py
  • in line 99 we added a function called load_data_rpForest that returns an adjacency matrix based on rpForest
• \GCN\train.py
  • in line 29 we called the function utils.load_data_rpForest to work on adjacency matrix based on rpForest

Files that we modified from the original LDS code

• \LDS\RPTree.py
  • a code that returns an rpTree given a feature matrix X
• \LDS\lds.py
  • in line 302 we added a code that returns an adjacency matrix based on rpForest
• \LDS\hyperparams.py
  • in line 177 we added a code that randomly picks a percentage of edges that were missed by rpForest

Setup

• Download and install Anaconda Navigator

• launch Anaconda Prompt and run the following commands:

  • conda create -n tf15 python tensorflow=1.15
  • conda activate tf15
  • conda remove --force tensorflow-estimator
  • conda install -c anaconda tensorflow-estimator==1.15.1
  • conda install -c anaconda scikit-learn
  • conda install -c conda-forge munkres
  • conda install -c conda-forge python-annoy
  • conda install -c conda-forge keras==2.3.1
  • conda install -c anaconda spyder

• To start working in this enviroment, launch Anaconda Prompt and type:

  • activate tf15
  • spyder