Baselines for JSSP

[cshjin]

• ACO (heuristic-based swarm algorithms)
• ACO_LS (our approach)
• OR-Tools (serve as ground truth, but should be considered as base line)
• RL (L2D, Jsp-env etc) reinforcement learning based algorithms

Metrics:

• Relative accuracy of minimum makespan (the ratio of makespan reached compared to ground truth)
• Wall-time of execution (including local search, RL updating etc)
• Scalability of instances, beyond the benchmark cases, generate instances
  • either synthetically from randomness or statistically from real systems

[cshjin]

https://github.com/swarm-workflows/aco-scheduling/tree/benchmark has the RL-based scheduler

[cshjin]

baselines

• greedy (list scheduling)
• binary variables
• mixed of permutation variables
• OR-Tools (CP-SAT)
• JSS
• ACO
• ACO_LS

cases

TA 41-51, add more cases from the benchmark folder if needed.