ai-for-decision-making-tue / job_shop_scheduling_benchmark_environments_and_instances

A benchmarking repo with various solution methods to various machine scheduling problems

Job Shop Scheduling Benchmark: Environments and Instances for Learning and Non-learning Methods

Welcome to the Job Shop Scheduling Benchmark

This GitHub repository serves as a comprehensive benchmark for a wide range of machine scheduling problems, including Job Shop Scheduling (JSP), Flow Shop Scheduling (FSP), Flexible Job Shop Scheduling (FJSP), FJSP with Assembly constraints (FAJSP), FJSP with Sequence-Dependent Setup Times (FJSP-SDST), and the online FJSP (with online job arrivals). Our primary goal is to provide a centralized hub for researchers, practitioners, and enthusiasts interested in tackling machine scheduling challenges.

🛠 Solution methods

We aim to include a wide range of solution methods capable of solving machine scheduling problems with various constraints and characteristics. This selection ranges from load-balancing heuristics, dispatching rules and genetic algorithms to end-to-end Deep Reinforcement Learning solutions. The repo currently contains the following solution methods, each capable of solving machine scheduling problems with the corresponding characteristics:

Solution methods	Job Shop (JSP)	Flow Show (FSP)	Flexible Job Shop (FJSP)	FJSP SDST	FAJSP	Online (F)JSP
Dispatching Rules	✓	✓	✓		✓	✓*
Genetic Algorithm	✓	✓	✓	✓	✓
MILP	✓	✓	✓	✓
CP-SAT	✓	✓	✓	✓
FJSP-DRL	✓	✓	✓
L2D	✓	✓
DANIEL	✓	✓	✓

*Capable of online arrivals of FJSP problems

🔜 We have a few DRL-based solutions in the pipeline, which will be published here upon completion.

📢 We encourage you to make use of our repository to get started with your own solutions, and, when possible, release your solution method in this repository.

📝 Cite our repository:

Please consider citing our paper if you use code or ideas from this project:

Robbert Reijnen, Kjell van Straaten, Zaharah Bukhsh, and Yingqian Zhang (2023) Job Shop Scheduling Benchmark: Environments and Instances for Learning and Non-learning Methods. arXiv preprint arXiv:2308.12794

@misc{reijnen2023job,
      title={Job Shop Scheduling Benchmark: Environments and Instances for Learning and Non-learning Methods}, 
      author={Robbert Reijnen and Kjell van Straaten and Zaharah Bukhsh and Yingqian Zhang},
      year={2023},
      eprint={2308.12794},
      archivePrefix={arXiv},
      primaryClass={cs.AI}
}

job_shop_scheduling_benchmark_environments_and_instances's Issues

Validation error when training DRL method with given configuration file and data.

Hi~ , I'm a newbee in FJSP, and I'm very appreciate for this project. When i try to run the training method of DRL solution, I encounted the following problem when validating.

I guess the validation dataset holds the wrong samples, but i don't know how to solve.
The traceback stack is as follows:

Exception has occurred: KeyError
2
File "/home/cheny/Projects/JSPRepo/solutions/FJSP_DRL/env.py", line 252, in step
duration = operation.processing_times[machine_ix]
~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^
File "/home/cheny/Projects/JSPRepo/solutions/FJSP_DRL/validate.py", line 48, in validate
state, rewards, dones = env.step(actions)
^^^^^^^^^^^^^^^^^
File "/home/cheny/Projects/JSPRepo/solutions/FJSP_DRL/train.py", line 175, in main
vali_result, vali_result_100 = validate(env_valid_paras, env_valid, model.policy_old)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/cheny/Projects/JSPRepo/solutions/FJSP_DRL/train.py", line 219, in
main(param_file=args.config_file)
KeyError: 2

And the keyError value may differ when re-run the training process.
Any helps would be appreciated.

你好，请问应该用哪个python版本呢？

我用py3.12的环境，有些包安装不了。

Hosting best-known solutions of benchmark instances

Hi all, just came across this library by coincidence. Great to see this effort to make a benchmark repository! I'm working on flexible job shop scheduling problems and flow shop problems for my own research.

One major difficulty I have is that most papers/repositories don't store their solutions but only report the best-found value. This makes it hard to verify their results. In vehicle routing, we have e.g., CVRPLIB which maintains the best-known solutions for CVRP and VRPTW. We also try to do that for prize-collecting VRP here.

Are there any plans from your side to host the best-known solutions? Happy to contribute as well.