python-orca

Orbit counting algorithm implemented by python.

Description

Orca is an efficient algorithm for counting graphlets in undirect and no-selfloop networks. It computes node- and edge-orbits (of 4- and 5-node graphlets) for each node in the network.

To improve the understanding of the orca for beginners, we re-implemented orca algorithm using python.

To improve the efficiency of python code, we use python-numba-cuda to accelerate orca on GPU.

Installation

To accelerate orca on GPU, we have three requirement packages: cupy, cudf, and numba.

conda create -n cupy-orca -c rapidsai -c nvidia -c conda-forge cudf=22.08 python=3.8 cudatoolkit=11.5
conda deactivate
conda activate cupy-orca
git clone https://github.com/xiangsheng1325/python-orca
cd python-orca

Usage

The utility takes four command-line arguments:

python orca.py --orbit-type node --graphlet-size 4 --input-file example_graph.csv --output-file orbit-counts.csv

Orbit type is either node or edge.
Graphlet size indicates the size of graphlets that you wish to count and should be either 4 or 5.
Input file describes the network in a simple text format. The first line contains two integers n and e - the number of nodes and edges. The following e lines describe undirected edges with space-separated ids of their endpoints. Node ids should be between 0 and n-1. See example_graph.csv as an example.
Output file will consist of n lines, one for each node in a graph from 0 to n-1. Every line will contain space-separated orbit counts depending on the specified graphlet size and orbit type.

Data

The random Erdos-Renyi graph used in the experiment is available in the example_graph.zip archive.

Comparisons with other implementations

Erdos-Renyi graph with 1M nodes and 16M edges

Time consumptions (seconds) comparison

Algorithm	Stage 1	Stage 2	Stage 3	Total
orca-c++	7.04	4.52	91.76	103.32
orca-python (Speedup)	0.64 (11.0$\times$)	0.19 (23.8$\times$)	2.87 (32.0$\times$)	3.70 (27.9$\times$)

xiangsheng1325 / python-orca