mpe_vdn.yaml

seed: 0
lr: 7e-4
episode_length: 200
num_mini_batch: 20
run_dir: ./run_results/
experiment_name: train_mpe_vdn
log_interval: 10
use_valuenorm: true
use_adv_normalize: true
wandb_entity: openrl-lab
callbacks:
  - id: "CheckpointCallback"
    args: {
        "save_freq": 500, # how often to save the model
        "save_path": "./results/checkpoints/",  # where to save the model
        "name_prefix": "vdn", # the prefix of the saved model
        "save_replay_buffer": True # not work yet
    }
  - id: "EvalCallback"
    args: {
      "eval_env": {"id": "simple_spread","env_num":1}, # how many envs to set up for evaluation
      "n_eval_episodes": 4, # how many episodes to run for each evaluation
      "eval_freq": 500, # how often to run evaluation
      "log_path": "./results/eval_log_path", # where to save the evaluation results
      "best_model_save_path": "./results/best_model/", # where to save the best model
      "deterministic": True, # whether to use deterministic action
      "render": False, # whether to render the env
      "asynchronous": True, # whether to run evaluation asynchronously
    }