kengz/SLM-Lab

{
  "dqn_per_qbert": {
    "agent": [{
      "name": "DQN",
      "algorithm": {
        "name": "DQN",
        "action_pdtype": "Argmax",
        "action_policy": "epsilon_greedy",
        "explore_var_spec": {
          "name": "linear_decay",
          "start_val": 1.0,
          "end_val": 0.01,
          "start_step": 10000,
          "end_step": 1000000
        },
        "gamma": 0.99,
        "training_batch_iter": 1,
        "training_iter": 4,
        "training_frequency": 4,
        "training_start_step": 10000
      },
      "memory": {
        "name": "PrioritizedReplay",
        "alpha": 0.6,
        "epsilon": 0.0001,
        "batch_size": 32,
        "max_size": 200000,
        "use_cer": false
      },
      "net": {
        "type": "ConvNet",
        "conv_hid_layers": [
          [32, 8, 4, 0, 1],
          [64, 4, 2, 0, 1],
          [64, 3, 1, 0, 1]
        ],
        "fc_hid_layers": [256],
        "hid_layers_activation": "relu",
        "init_fn": null,
        "batch_norm": false,
        "clip_grad_val": 10.0,
        "loss_spec": {
          "name": "SmoothL1Loss"
        },
        "optim_spec": {
          "name": "Adam",
          "lr": 2.5e-5,
        },
        "lr_scheduler_spec": null,
        "update_type": "replace",
        "update_frequency": 1000,
        "gpu": true
      }
    }],
    "env": [{
      "name": "QbertNoFrameskip-v4",
      "frame_op": "concat",
      "frame_op_len": 4,
      "reward_scale": "sign",
      "num_envs": 16,
      "max_t": null,
      "max_frame": 4e6
    }],
    "body": {
      "product": "outer",
      "num": 1
    },
    "meta": {
      "distributed": false,
      "eval_frequency": 10000,
      "log_frequency": 10000,
      "rigorous_eval": 0,
      "max_session": 4,
      "max_trial": 1
    }
  }
}