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')

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

dynamicFeatures(mutag,lmutag,1,3)

3. 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])

4. Training classifier. ex:

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

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