pyoslom 0.1.3

OSLOM graph clustering algorithm

Python binding for OSLOM graph clustering algorithm

Summary

Pyolsom is a python binding for OSLOM (Order Statistics Local Optimization Method) graph clustering algorithm.

It works with directed/undirected weighted and unweighted graph. The algorithm performs usually good but slow, so it is better to be applied to medium graph size.

The orginal C++ code is really hard to be refactored. I tried the best to make it work with python.

Known issues

• The lib is not thread safe. So use mutliprocess when parallel is required.

Requirements

• C++ 17
• Python 3
• scikit-learn>=0.24
• pybind11>=2.6
• networkx>=2.5

The versions are what I worked on. Lower versions may work also.

Install

Install pybind11 first because there is no binary release in pip repo and the setup.py depends on pybind11.

pip install "pybind11>=2.6"

On Windows install Microsoft Visual C++ Build Tool first (refer to https://wiki.python.org/moin/WindowsCompilers).

build from source

git clone https://[e-mail]/bochen0909/pyoslom.git && cd pyoslom && python setup.py install

or use pip

pip install pyoslom

How to use

Example:

import networkx as nx
from pyoslom import OSLOM

G = nx.read_pajek("example.pajek") # networkx graph or adjacency matrix
alg = OSLOM(random_state=123)
results = alg.fit_transform(G)

def print_clus(clus):
    for k, v in clus.items():
        if k != 'clusters':
            print(str(k) + "=" + str(v))
    for k, l in clus['clusters'].items():
        print("Level:" + str(k) + ", #clu=" + str(len(l)))

print_clus(results)

For more complete examples please see the notebook example.ipynb.

License

The original c++ code is published at OSLOM following a research publication. However there is no license attached with it. The python wrapping work is licensed under the GPLv2.