/Dynfeats

Dynamics Based Features For Graphs Classification

Primary LanguageMATLAB

Dynfeats

This repository provides sources of the method proposed in the paper: Dynamics Based Features for Graph Classification. For further information, please check the scripts header.

This is a feature based method for creating graph feature vectors for classification purposes. More details can be found in the website.

Scripts for svm are available in Matlab. The Random Forest version is in Python.

USAGE:

  1. Loading dataset. ex:

load('datasets/mutag.mat')

  1. Generating features. This script will generate all features. ex:

dynamicFeatures(mutag,lmutag,1,3)

  1. Selecting features acording with the following codification:

             1 - H = identity
         2 - pagerank
         3 - second eigenvector
         4 - node labels encoded in H
         5 - number of nodes
         6 - number of edges
         7 - Nodes degrees
         8 - Node betweeness
         9 - Local clustering coefficient
         10 - Closeness centrality        
         11 - Degree centrality
         12 - Assortativity
         13 - Number of triangles
         14 - Global clustering coefficient

ex. feats = load_dynfeat([1 4])

  1. Training classifier. ex:

result=runntimes(feats, lmutag,10,1,0)

Note: Do not forget to change your libsvm path into runIndependent_dynF.m file.