GMS

GMS is a graph neural network model which can learn to predict the solution of MaxSAT problem, an optimization variant of the well-known Boolean Satisfiability problem (SAT).

Usage

We published the sourse code of GMS, as well as the problem generation, training and testing scripts.

1. Clone this repo to your machine

	git clone https://github.com/minghao-liu/GMS.git

We only tested this project on a server with NVIDIA Tesla V100 GPU, and PyTorch 1.5.0.

2. Setup

	chmod 755 *.sh
	./setup.sh

Create the required folders, and download the problem generator [link], as well as the baseline MaxSAT solvers we use: MaxHS, Loandra and SATLike.

3. Generate raw data

	 ./generate_raw_data.sh

The generated problems are in the standard DIMACS format.

4. Generate data

	 ./generate_data.sh

The data files will be stored in the data folder, which can be loaded by GMS.

5. Train the GMS models

	 ./train.sh

The trained model will be stored in the model folder.