Graph Analysis Toolbox

The Graph Analysis Toolbox is a generic software for the management and processing of a interrelated collection of multiple graphs.

It can be used to process multiple graphs. Functionality include (but it is not limited to):

1. Similarity graphs: generated from node attributes, based on different similarity measures (Jensen-Shannon, Hellinger, L1, L2).
   • General implementations based on the neighbors module from scikit-learn.
   • Specific implementation for fast computation of Hellinger distances using Numba and cuda.
2. Community detection algorithms (Louvain, Walktrap, FastGreeedy, Label Propagation)
   • Implementations based on IGraph and Networkx.
3. Bipartite graphs from attributes
4. Transductive graphs: Graphs generated by connecting target nodes from a bipartite graph. Link weights are computed from the links of a graph connecting the source nodes.
5. Transitive graphs, computed as the composition of two bipartite graphs.
6. Analysis of graph partitions.
7. Analysis of graph nodes (centrality measures, PageRank).
   • Implementations based on Networkx.
8. Edicion tools for the collection of graphs:
   • Create, add, remove graphs
   • Subsampling
   • Reduction to graphs of equivalence classes
9. Tools for visualization:
   • Graph layout algorithms.
   • Exportation to GEXF format
   • Visualization of bipartite graphs (requires Halo, not included)

Usage:

As an application:

The software includes two applications that can be used to generate and manipulate graphs through an interactive menu:

• mainRDIgraphs.py: Provides accces to the sofware functionality through an interative menu. It reads the links to the source data from a configuration file (parameters.yaml). You would need to edit this file to use other data.
• mainRDIlab.py: It uses the software functionality to carry out experiments for analysing RDI corpus collections.

Write

python mainRDIgraphs.py --h
python mainRDIlab.py --h

to see the available options.

As a sofware package:

The software include several class packages that can be used independently. Classes include (and are not limited to):

• SimGraph: Generation of similarity graphs
• CommunityPlus: Wrapper to community detection algorithms
• DataGraph (requires SimGraph and CommunityPlus): provides tools for graph processing and analysis.
• SuperGraph (requires DataGraph): provides tools for handling collections of DataGraph objects, including tools for the generation of new datagraphs.