kengz/SLM-Lab

{
  "ppo_lunar": {
    "agent": [{
      "name": "PPO",
      "algorithm": {
        "name": "PPO",
        "action_pdtype": "default",
        "action_policy": "default",
        "explore_var_spec": null,
        "gamma": 0.99,
        "lam": 0.95,
        "clip_eps_spec": {
          "name": "no_decay",
          "start_val": 0.20,
          "end_val": 0.0,
          "start_step": 10000,
          "end_step": 300000
        },
        "entropy_coef_spec": {
          "name": "no_decay",
          "start_val": 0.01,
          "end_val": 0.01,
          "start_step": 0,
          "end_step": 0
        },
        "val_loss_coef": 1.0,
        "time_horizon": 128,
        "minibatch_size": 256,
        "training_epoch": 10
      },
      "memory": {
        "name": "OnPolicyBatchReplay",
      },
      "net": {
        "type": "MLPNet",
        "shared": false,
        "hid_layers": [64, 64, 32],
        "hid_layers_activation": "relu",
        "init_fn": "orthogonal_",
        "batch_norm": false,
        "clip_grad_val": 0.5,
        "use_same_optim": true,
        "loss_spec": {
          "name": "MSELoss"
        },
        "actor_optim_spec": {
          "name": "Adam",
          "lr": 5e-4,
        },
        "critic_optim_spec": {
          "name": "Adam",
          "lr": 5e-4,
        },
        "lr_scheduler_spec": null,
        "gpu": false
      }
    }],
    "env": [{
      "name": "LunarLander-v2",
      "frame_op": "concat",
      "frame_op_len": 4,
      "max_t": null,
      "max_frame": 300000,
      "num_envs": 8,
      "normalize_state": false
    }],
    "body": {
      "product": "outer",
      "num": 1
    },
    "meta": {
      "distributed": false,
      "log_frequency": 1000,
      "eval_frequency": 1000,
      "max_session": 4,
      "max_trial": 1
    }
  },
}