ftr_algo_ddpg_cfg.yaml

params:
  seed: ${.config.seed}

  algo:
    name: ddpg

  load_checkpoint: ${if:${.config.checkpoint},True,False} # flag which sets whether to load the checkpoint
  load_path: ${.config.checkpoint} # path to the checkpoint to load

  config:
    seed: 40
    name: ftr_algo_ddpg
    experiment: ""
    device: cuda:0
    max_iterations: ""
    checkpoint: ""
    test: False

  learn:
    agent_name: ${..config.name}
    full_experiment_name: logs/${resolve_default:${..config.name},${..config.experiment}}
    device: ${..config.device}
    device_name: ${..config.device}

    test: ${..config.test}

    resume: 0
    save_interval: 500 # check for potential saves every this many iterations
    print_log: True

    # rollout params
    max_iterations: 6000

    # training params
    hidden_nodes: 512
    hidden_layer: 3

    cliprange: 0.2
    nsteps: 8
    noptepochs: 5
    nminibatches: 4 # this is per agent #TODO: too small
    replay_size: 5000
    polyak: 0.995
    learning_rate: 0.0001
    max_grad_norm: 1
    gamma: 0.99
    act_noise: 0.1
    target_noise: 0.2
    noise_clip: 0.5
    reward_scale: 1
    batch_size: 64

    log_interval: 1
    asymmetric: False