[Research Note] Hyperparameter optimization

[xoxown31]

Intro

In this discussion, let's talk about hyperparameter optimization for RQ1.
We will find and organize the hyperparameters that RL can best learn in each situation.

Method

Perform various experiments by modifying player_configs on optimize.py.
After that, you can write a comment on this discussion page according to the form below.

Template

- Experimental information
RL {int} + RBA {int}
RL 1: {Role}
RL 2: {Role}
...

- Optimal hyperparameters
...

[chang1025]

• Experimental information
  RL {2} + RBA {6}
  RL 1: {Mafia}
  RL 2: {Mafia}

• Optimal hyperparameters
  {'lr': 0.00029291989456433845, 'gamma': 0.9700125598295506, 'entropy_coef': 0.015169565812093458, 'batch_size': 4096, 'hidden_dim': 512, 'eps_clip': 0.01}

[xoxown31]

• Experimental information
  RL 1 + RBA 7
  RL 1: Police

• Optimal hyperparameters
  {'lr': 0.000013798316269031359, 'gamma': 0.9838236614730617, 'entropy_coef': 0.020675525613775662, 'batch_size': 2048, 'hidden_dim': 256, 'eps_clip': 0.02}

[chang1025]

• Experimental information
  RL 1 + RBA 7
  RL 1: Doctor

• Optimal hyperparameters
  {'lr': 0.0002799871885499148, 'gamma': 0.9027856916043727, 'entropy_coef': 0.040274904708271206, 'batch_size': 4096, 'hidden_dim': 512, 'eps_clip': 0.01}

[chang1025]

• Experimental information
  RL 4 + RBA 4
  RL 1: Citizen
  RL 2: Citizen
  RL 3: Citizen
  RL 4: Citizen

• Optimal hyperparameters
  {'lr': 0.0001145641033340477, 'gamma': 0.9987590799611283, 'entropy_coef': 0.015388190250989421, 'batch_size': 4096, 'hidden_dim': 256, 'eps_clip': 0.01}