2022-02-08 meeting overview

[MichalisPanayides]

What has been done:

• Machine Learning course:
  • Week 1-4: ✔️
  • Week 5: ⌚
• Simulation compatibility with game ⌚
• Work on RL algorithm: ⌚
  • Utility functions
  • Random server allocation
  • Reinforcement learning implementation
  • Some early results

To be discussed:

• [x] Researchfish annual review
• [x] Plugged in distributions from paper: Server behaviours in healthcare queueing systems
• [x] Reinforcement Learning
  • first implementation and some results
• [x] Decisions:
  • [x] Utility function - (Need to choose one that would make sense)
  • [x] Server priority?
  • [x] Initial rates?
  • [x] Reinforcement learning before the actual game??

To be done:

• Researchfish review
• Continue RL algorithm
• Continue ML course
• Continue working on thesis structure
• Abstract for EURO2022
• Abstract for CORS/INFORMS

[MichalisPanayides]

• [ ] Distractions, Risks, Recoveries -> Create vague abstract so that we can incorporate RL extension if needed
• [x] EURO 2022 -> Email with ORAHS code
• [ ] Researchfish submission
• [x] Minor things to change:
  • [x] Parametrisation.ipynb: (Arrival time * Servers) instead of (Service time / Servers)
  • [x] Use utilisation instead of busy time
  • [x] Use average simulated service time instead of service rate
• [x] Reinforcement Learning:
  • [x] Improve function names and logic to be more RL oriented
  • [x] Decrement, Increment -> pick_policy

    pick_policy()
    run_simulation()
    if policy_improved_utility():
    accept_policy()

  • [x] Plot some form of overall policy next to utility graph
  • [x] Add a limit on neighborhood size
  • [x] Increase number of trials in actual simulation
• [ ] Future things to keep in mind:
  • When service rate at a state is changed, change the surrounding states
  • Look at custom server priority
• [ ] Game paper major revisions 😭