behaviors:
  3DBallHard:
    trainer_type: ppo_transfer
    hyperparameters:
      batch_size: 1200
      buffer_size: 12000
      learning_rate: 0.0003
      beta: 0.001
      epsilon: 0.2
      lambd: 0.95
      num_epoch: 3
      learning_rate_schedule: linear
      model_schedule: linear
      # separate_value_train: true
      encoder_layers: 1
      policy_layers: 1
      forward_layers: 1
      value_layers: 1
      feature_size: 16
      reuse_encoder: true
      in_epoch_alter: false
      in_batch_alter: false
      use_op_buffer: false
      use_var_predict: true
      with_prior: false
      predict_return: true
      use_bisim: false
      separate_value_net: false
      use_transfer: true
      transfer_path: "results/ppomodel_ball_l1/3DBall"
      load_model: true
      train_model: false
      # load_policy: true
      # load_value: true
    network_settings:
      normalize: true
      hidden_units: 128
      num_layers: 2
      vis_encode_type: simple
    reward_signals:
      extrinsic:
        gamma: 0.995
        strength: 1.0
    keep_checkpoints: 5
    max_steps: 2000000
    time_horizon: 1000
    summary_freq: 12000
    threaded: true