kpolimis / graphreduce

Graphreduce

Large scale network reduction

Graphreduce finds hierarchical community structure (compression patterns) in large complex networks.

The code here attempts to be pretty lightweight and straightforward, the heavy lifting is done by:

• GraphLab
• The map equation
• Relaxmap

Some details

Graphreduce was built to squeeze the essence out of large complex networks. It can find hierarchical structure in any directed weighted net.

We owe a lot to GraphLab, a free data analysis framework aimed at enabling web scale machine learning and network analysis.

We're also indebted to the work done by Martin Rosvall and map equation team. The map equation is an information theory centered objective function used to quantify pattern detection on networks. Relaxmap is a parallel implementation of the map equation objective. Dare we not mention the father of information theory Claude Shannon.

Try it

$ git clone https://github.com/timmytw/graphreduce.git

$ cd graphreduce/; pip install -r requirements.txt

$ python graphreduce/test/test.py

Run it on your own network

$ python graphreduce/reduce.py vertex_path edge_path output_dir

See it in action

Graphreduce is a key part of SmartTypes.org, a free web service aimed at social exploration.

Get in touch @ [email protected]