bkj / sgm

SGM in Pytorch

sgm

Python port of seeded graph matching (SGM)

Quickstart

# make fresh conda env
conda create -n sgm_env python=3.6 pip -y
source activate sgm_env

# install requirements
cat requirements.txt | xargs -I {} pip install {}

# install sgm
pip install -e .

Tested on Ubuntu 16.04. Installation on OSX may fail.

Example Usage

Simple example:

cd examples/synthetic
python main.py

More complex example:

cd examples/kasios/
python kasios.py --num-nodes 1000 --num-seeds 32 --backend scipy.classic.jv

sgm's Issues

Sparse IO

Right now, we're reading graphs as dense adjacency matrices, which is a terrible idea. Need to implement sparse IO functions.