seed: null # The seed to generate the world. If null, the seed will be randomly generated.
env_path: null # The path to the MinecraftEnv project. If null, it will use the default path.
group: jsrl_fish # The name of the group of the experiment. The experiments will be grouped by this name in the wandb console.
record_video: true # Whether to record the video of the experiment.
device: null # The device to run the experiment. If null, it will use the default device.

env: # The environment to run the experiment in.
  name: # The name of the pre-defined environment. You can choose one from the list in environments.py
    params: # The parameters to create the environment. You can see the parameters in environments.py
      hud: true # Whether to show the HUD in the environment.
      verbose: false # Whether to show the verbose information in the environment.
      port: 8000 # Via which port the environment will communicate with this agent.
      render_action: true # Whether to include the agent's action in the video.
      size_x: 256 # The size of the screen in x axis.
      size_y: 256 # The size of the screen in y axis.

wrappers: # The wrappers that wraps the environment. It is an array of wrappers that will be applied in order.
  - name: 'ActionWrapper' # The convenience wrapper for the discrete actions.
    enabled_actions: # The specific actions that are enabled by this wrapper.
      - NO_OP # No operation action.
      - USE # Use action.

  - name: 'SoundWrapper' # The wrapper that adds sound-based features to the environment.
    coord_dim: 2 # The dimension of sound coordinates.
  - name: 'FishCodWrapper' # Additional wrapper for fishing reward.

algorithm: # Details of the algorithm to be used.
  name: "SoundJSRLDQNAlgorithm" # The specific name of the algorithm.
  params: # Parameters for the algorithm.
    num_episodes: 1000 # Total number of episodes to run.
    warmup_episodes: 10 # Episodes before starting the main training.
    steps_per_episode: 400 # Steps to be taken per episode.
    test_frequency: 10 # Frequency of tests during the episodes.
    solved_criterion: # Criterion to determine if the environment is solved.
      name: 'ScoreCriterion'
      params:
        min_episode: 100
        min_avg_score: 195
        min_test_score: 195
        min_avg_test_score: 195
  hidden_dim: 128 # Dimension of the hidden layers.
  epsilon_init: 1.0 # Initial value of epsilon for the epsilon-greedy policy.
  epsilon_decay: 0.99 # Decay rate of epsilon.
  epsilon_min: 0.01 # Minimum value of epsilon.
  update_frequency: 1000 # Frequency of updates.
  train_frequency: 1 # Training frequency.
  replay_buffer_size: 1000000 # Size of the replay buffer.
  batch_size: 256 # Batch size for training.
  gamma: 0.99 # Discount factor.
  learning_rate: 0.00001 # Learning rate for the optimizer.
  weight_decay: 0.00001 # Weight decay for regularization.
  tau: 1.0 # Target network update rate.
  guide_policy: # Policy that guides the agent.
    name: 'FishingGuide'
    params:
      min_episode: 100
  decrease_guide_step_threshold: 0.5 # Threshold for decreasing guide steps.