NetRL: Task-aware Network Denoising via Deep Reinforcement Learning

About

This project implements the NetRL model, a deep reinforcement learning based method for network denoising, which can be guided by downstream task.

Dependencies

The script has been tested running under Python 3.5.2, with the following packages installed (along with their dependencies):

• tensorflow==1.14.0
• numpy==1.17.0
• pandas==0.24.2
• sklearn==0.21.3
• scipy==1.3.1
• tqdm==4.32.1

Some Python module dependencies are listed in requirements.txt, which can be easily installed with pip:

pip install -r requirements.txt

In addition, CUDA 9.0 has been used in our project. Although not all dependencies are mentioned in the installation instruction links above, you can find most of the libraries in the package repository of a regular Linux distribution.

Input Format

Some example data formats are given in data for your reference. The input files are expected to be three parts:

1. Dataset details, where x is the node feature vectors, y is the one-hot labels and graph is a dict in the format {index: [index_of_neighbor_nodes]}, where the neighbor nodes are organized as a list. In this example, we use the preprocessed citation network Cora and Citeseer provided by https://github.com/kimiyoung/planetoid (Zhilin Yang, William W. Cohen, Ruslan Salakhutdinov, Revisiting Semi-Supervised Learning with Graph Embeddings, ICML 2016).

2. edgelist is the network edgelist with different noise ratio, e.g.,

    1 2
    1 3
    1 4
    ...
   

3. embedding is the pretrained node representations where the first line is header (# of nodes, # of dimension), all other lines are node-id and d dimensional representation:

    32 64
    1 0.016579 -0.033659 0.342167 -0.046998 ...
    2 -0.007003 0.265891 -0.351422 0.043923 ...
    ...
   

How to run NetRL

The help information of the main script main.py is listed as follows:

python Main.py -h

usage: Main.py [-h][--env_name] [--use_gpu] [--gpu_id] [--gpu_fraction] [--random_seed]

optional arguments:
  -h, --help                show this help message and exit
  --env_name                str, select the dataset. 
  --use_gpu                 bool, whether to use gpu or not.
  --gpu_id                  str, which gpu to use.
  --gpu_fraction            str, idx / # of gpu fraction e.g. 1/3, 2/3, 3/3
  --random_seed             int, value of random seed

We set the default parameters in config.py. You can modify them when using your own dataset.