https://github.com/diyang/GRAPH-NODE-RECOGINITION-R

MxNet implementation of Node Recognition in Large Scale Network
https://github.com/diyang/GRAPH-NODE-RECOGINITION-R

Last synced: 2 months ago
JSON representation

MxNet implementation of Node Recognition in Large Scale Network

• Host: GitHub
• URL: https://github.com/diyang/GRAPH-NODE-RECOGINITION-R
• Owner: diyang
• License: gpl-3.0
• Created: 2018-11-21T22:09:06.000Z (about 6 years ago)
• Default Branch: master
• Last Pushed: 2018-12-09T13:17:09.000Z (about 6 years ago)
• Last Synced: 2024-05-22T04:19:52.484Z (8 months ago)
• Language: R
• Homepage:
• Size: 9.25 MB
• Stars: 3
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• Awesome-MXNet - SAGE-GRAPH

README

        
# Node Recognition in Large Scale Network

This repo contains an MXNet implementation of [this](https://arxiv.org/pdf/1706.02216.pdf) state of the art Graph Convolutional Neural Network with R.

![](./docs/sample_and_agg.png)

## Requirements

All examples use MxNet 0.10.1 library to perform deep learning, 

and use igraph 1.2.2 to store graph regarding data. To install the library, please type following commands in R Studio

  

```

install.package("mxnet") 

install.package("igraph")

```

## Running the code

Download & extract the training data:

Currently the examples in this directory are tested on the CORA dataset. The GraphSAGE model assumes that node

features are available.

The dataset can be found in example_data folder, and data are in CSV format

The following is the description of the dataset:

> The Cora dataset consists of 2708 scientific publications classified into one of seven classes.

> The citation network consists of 5429 links. Each publication in the dataset is described by a

> 0/1-valued word vector indicating the absence/presence of the corresponding word from the dictionary.

> The dictionary consists of 1433 unique words. The README file in the dataset provides more details.

## Results & Comparison

- Please run `main.R`

- This MXNet implementation achieves *NLL = 0.780* after 100 epochs on the validation dataset

## Hyperparameters

The default arguments in `main.R` achieve equivalent performance to the published results. For other datasets, the following hyper parameters provide a good starting point:

- K = 2

- hidden num  = {20,20}

- random neighbour sampling = {20,10}

- learning rate = 0.005

- Dropout after every layer =  0.3

- Epochs = 100+