eqmarl/experiments/coingame_maa2c_mdp_sctde.yml at main · news-vt/eqmarl · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
experiment:

  roots:
    root_dir: &root_dir '../experiment_output/coingame_maa2c_mdp_sctde'
    session_dir: &session_dir !join [*root_dir, '/', '{datetime_session:%Y%m%dT%H%M%S}']
    checkpoint_root: !join [*root_dir, '/', 'checkpoints']

  # Filenames to use for saving results at the end of training.
  save:
    metrics_file: &metrics_file !join [*session_dir, '/', 'metrics-{round}.json']
    model_files:
      - name: &actor_name "actor-classical-shared"
        filepath: &actor_file !join [*session_dir, '/', 'actor-{round}.weights.h5']
        save_weights_only: &actor_weight_only_flag True
      - name: &critic_name "critic-classical-joint"
        filepath: &critic_file !join [*session_dir, '/', 'critic-{round}.weights.h5']
        save_weights_only: &critic_weight_only_flag True

  # Training parameters.
  train:
    n_episodes: 3000
    callbacks:
      - func: eqmarl.AlgorithmResultCheckpoint
        params:
          filepath: *metrics_file
          save_freq: 100
          verbose: True
      - func: eqmarl.AlgorithmModelCheckpoint
        params:
          model_name: *actor_name
          filepath: *actor_file
          save_weights_only: *actor_weight_only_flag
          save_freq: 100
          verbose: True
      - func: eqmarl.AlgorithmModelCheckpoint
        params:
          model_name: *critic_name
          filepath: *critic_file
          save_weights_only: *critic_weight_only_flag
          save_freq: 100
          verbose: True

  # Plotting parameters.
  plot:
    plotargs:
      plot_data: mean
      # error_method: minmax
      error_method: std
    figsize: [10,8]
    mosaic: [[undiscounted_reward, coins_collected], [own_coins_collected, own_coin_rate]]
    axes:
      undiscounted_reward:
        title: Score
        xlabel: epoch
        ylabel: value
      coins_collected:
        title: Coins Collected
        xlabel: epoch
        ylabel: value
      own_coins_collected:
        title: Own Coins Collected
        xlabel: epoch
        ylabel: value
      own_coin_rate:
        title: Own Coin Rate
        xlabel: epoch
        ylabel: value

  # The algorithm to run.
  # The keys within the `init_params` key are used with direct substitution in the class initializer.
  # This includes the following definitions:
  # - Environment
  # - Models
  # - Optimizers
  algorithm:

    init_func: eqmarl.algorithms.MAA2C
    init_params:
      gamma: 0.99
      alpha: 0.001
      episode_metrics_callback: eqmarl.environments.coin_game.episode_metrics_callback

      env:
        func: eqmarl.environments.coin_game.vector_coin_game_make
        params:
          domain_name: CoinGame-2
          gamma: 0.99
          time_limit: 50

      model_actor:
        init_func: eqmarl.models.generate_model_CoinGame2_actor_classical_shared_mdp
        init_params:
          n_actions: 4
          units:
            - 12
          name: *actor_name
        build_shape: [null, 36]

      optimizer_actor:
        func: tensorflow.keras.optimizers.Adam
        params:
          learning_rate: 1.0e-3

      model_critic:
        init_func: eqmarl.models.generate_model_CoinGame2_critic_classical_joint_mdp
        init_params:
          n_agents: 2
          units:
            - 12
          name: *critic_name
        build_shape: [null, 2, 36]

      optimizer_critic:
        func: tensorflow.keras.optimizers.Adam
        params:
          learning_rate: 1.0e-3