pong_impala_speedup.yaml

alg_para:
  alg_name: IMPALAOpt

  alg_config:
    train_per_checkpoint: 3
    prepare_times_per_train: 4
    BATCH_SIZE: 1000

env_para:
  env_name: VectorAtariEnv
  env_info:
    name: PongNoFrameskip-v4
    vision: False
    dim: 42
    vector_env_size: 5

agent_para:
  agent_name: AtariImpalaOpt
  agent_num : 1
  agent_config:
    max_steps: 50
    complete_step: 10020000
    sync_model_interval: 2

model_para:
  actor:
    model_name: ImpalaCnnOpt
    state_dim: [42,42,4]
    input_dtype: uint8
    state_mean: 128.0
    state_std: 128.0
    action_dim: 6
    model_config:
      LR: 0.001
      sample_batch_step: 50  # default set to max_steps
      grad_norm_clip: 40.0


env_num: 50
speedup: False