IN GOD WE TRUST.

Home
Publications
>>NgramCNN<<
Projects
Dataset&Tools
Courses
Contact
Open Datasets!



Tag Cloud


ClustrMaps




  Intro

NgramCNN

NgramCNN is a deep convolution neural network tool for graph classification. (TKDE paper )

The core NgramCNN library is written in python and optimized supporting GPU with better performance.

Key Features

NgramCNN consists of three novel components.

First, we introduce the concept of n-gram block to transform each raw graph object into a sequence of n-gram blocks connected through overlapping regions.

Second, we introduce a diagonal convolution step to extract local patterns and connectivity features hidden in these n-gram blocks by performing n-gram normalization.

Finally, we develop deeper global patterns based on the local patterns and the ways that they respond to overlapping regions by building a n-gram deep learning model using convolutional neural network.


  Download (Tensorflow Version)

Latest Version

Download NgramCNN HERE (TF version).

The latest code is also available from the github source

Python 2.7 is required for NgramCNN.

Besides, following libs are needed:

-chainer

-numpy

-itertools-recipes

-futures

-six


  Download (Chainer Version)

Latest Version

Download NgramCNN HERE (Chainer version).

Enverionment

Python 2.7 is required for NgramCNN.

Besides, following libs are needed:

-chainer

-numpy

-itertools-recipes

-futures

-six

Both these libs can be downloaded from their website (just google them) or installed by pip.

$pip install numpy futures itertools_recipes six chainer 

DEMO

NgramCNN can be executed in demo mode by

$python ngramcnn_tuning.py

Note that we developed and tested these codes in MacOS and Ubuntu.

Windows OS may not support some OS commands and you can just remove those codes.

In default, the GPU is required.

Nvidia GeForce 1080 and titanX are suggested configuration.

If you want to try the CPU mode, even though not suggested, you can set the gpu=False in ngramcnn.py


  Cite

Please cite our publication if you'd like to use our code (for comparison and promotion).

Zhiling Luo, Ling Liu, Jianwei Yin, Ying Li, Zhaohui Wu (2017). Deep Learning of Graphs with Ngram Convolutional Neural Networks[J] IEEE Transactions on Knowledge and Data Engineering.

Here is the bib format:

@article{luo2017deep, title={Deep Learning of Graphs with Ngram Convolutional Neural Networks}, author={Luo, Zhiling and Liu, Ling and Yin, Jianwei and Li, Ying and Wu, Zhaohui}, journal={IEEE Transactions on Knowledge and Data Engineering}, year={2017}, publisher={IEEE} }



Last change: July 20, 2017