yunluoxiaobin / gnn_jssp

Master's Thesis - Graph Neural Networks for Compact Representation for Job Shop Scheduling Problems: A Comparative Benchmark

gnn_jssp

Master's Thesis - Matthew Judijanto - RWTH Aachen 2021

Title: Graph Neural Networks for Compact Representation for Job Shop Scheduling Problems: A Comparative Benchmark

Dependecies

1. PyTorch Geometric
2. PyTorch 1.7.0

Dataset

Datasets of this thesis are generated based on:

1. Normal datasets: https://www.scitepress.org/Link.aspx?doi=10.5220/0010202405890597
2. L2D datasets: https://arxiv.org/abs/2010.12367

Note: Dataset for problem above the size of 10x10 are removed in PyG form due to GitHub 50mb file size limit.

Models

• Baseline: -- MLP MLP.py -- Ensemble Boosting RunEnsembleBoosting.py

• GNN Models: -- GCN GCN.py -- GIN-0 GIN.py -- PNA PNA.py -- EGC EGC.py

To Run

To run and train the models, there are 2 ways to do this. For a simple single run, you can use RunModel.py. For running a K-Fold Cross Validation use cv_split_test.py

Others

DataExtractor.py DataExtractorNorm.py DisjunctiveGraph.py DisjunctiveGraph_v2.py are used to convert the raw dataset from JSON to PyG graph data.

loadDataset.py is to load the PyG graph datasets to be runned on the model