gnnparitygames
Project for graph neural networks learning to play parity games.
Science Score: 44.0%
This score indicates how likely this project is to be science-related based on various indicators:
-
✓CITATION.cff file
Found CITATION.cff file -
✓codemeta.json file
Found codemeta.json file -
✓.zenodo.json file
Found .zenodo.json file -
○DOI references
-
○Academic publication links
-
○Academic email domains
-
○Institutional organization owner
-
○JOSS paper metadata
-
○Scientific vocabulary similarity
Low similarity (7.9%) to scientific vocabulary
Repository
Project for graph neural networks learning to play parity games.
Basic Info
Statistics
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
- Releases: 0
Metadata Files
README.md
gnnprsolver
Requirements
pip install --user torch torch-geometric
Train models
Use
./gnn-pg-solver.py train --network GAT --output GAT_weights.pth games/game_1.txt solutions/solution_1.txt
or its equivalent using shorthand options
./gnn-pg-solver.py train -n GAT -o GAT_weights.pth games/game_1.txt solutions/solution_1.txt
Predict winning regions
We assume that the directory games contains a set of plain-text files containing parity games.
Use
./gnn-pg-solver.py predict --network GAT --weights GAT_weights.pth --output results.csv games/*
or its equivalent using shorthand options
./gnn-pg-solver.py predict -n GAT -w GAT_weights.pth -o results.csv games/*
Evaluate predictions
First, generate some predictions using the command in the previous section.
We assume that for each game games/game_XXXX.txt there exists a solution solutions/solution_game_XXXX.txt.
The evaluate subcommand expects the input to be in the form game prediction reference, so we need to add the final column to the results before handing them to the evaluation.
sed 's:^games/game_\([0-9]*\).txt.*$:solutions/solution_game_\1.txt:' < results.csv > solutions.csv
paste -d' ' results.csv solutions.csv | ./gnn-pg-solver.py evaluate
Owner
- Name: dwardy
- Login: dillon-ward
- Kind: user
- Repositories: 2
- Profile: https://github.com/dillon-ward
Passionate programmer currently studying Comp Sci at uni.
Citation (CITATION.cff)
# SPDX-FileCopyrightText: 2022 German Aerospace Center (DLR)
# SPDX-FileContributor: Alexander Weinert <alexander.weinert@dlr.de>
#
# SPDX-License-Identifier: CC-BY-NC-ND-3.0
cff-version: 1.2.0
message: "If you use gnn_pr_solver in your research, please cite it using these metadata."
title: gnn_pr_solver
abstract: "gnn_pr_solver is a prototypical incomplete solver for parity games based that uses graph neural networks"
authors:
- family-names: Hecking
given-names: Tobias
affiliation: "German Aerospace Center (DLR)"
orcid: "https://orcid.org/0000-0003-0833-7989"
license: MIT
repository-code: "https://github.com/DLR-SC/gnn_pr_solver"