11‐07‐2023 Weekly Tag Up
- Joe
- Chi-Hui
- 20 round experiment complete (on 4 agent env)
- All hyperparameters left same as what was used for paper
- Results very similar to what we saw in paper
- Online rollouts show similar G1 performance to the threshold policy (the g1 policy)
- Online rollouts show G2 performance that's slightly better than the queue policy (the g2 policy)
- No real trends evident in either case
- Ask Joewie about how many rounds to train 9 & 16 agent env
- Probably need 10 because it's more common in academia
- Expected to hear back about paper around November 30th
- Probably will have feedback regarding proof
- Experiments (from last week) are the other thing to be preparing
- Set up meeting with Joewie to discuss how to best prepare for review feedback
