kengz/SLM-Lab

View on GitHub
slm_lab/spec/experimental/dqn/dqn_lunar_search.json

Summary

Maintainability
Test Coverage
{
  "dqn_concat_polyak_lunar": {
    "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": 0.08,
          "start_step": 0,
          "end_step": 12000
        },
        "gamma": 0.99,
        "training_batch_iter": 3,
        "training_iter": 4,
        "training_frequency": 4,
        "training_start_step": 32,
        "normalize_state": true,
      },
      "memory": {
        "name": "Replay",
        "batch_size": 32,
        "max_size": 100000,
        "use_cer": false
      },
      "net": {
        "type": "MLPNet",
        "hid_layers": [
          400,
          200
        ],
        "hid_layers_activation": "relu",
        "clip_grad_val": null,
        "loss_spec": {
          "name": "MSELoss"
        },
        "optim_spec": {
          "name": "Adam",
          "lr": 0.002
        },
        "lr_scheduler_spec": {
          "name": "MultiStepLR",
          "milestones": [
            80033
          ],
          "gamma": 0.1
        },
        "update_type": "polyak",
        "polyak_coef": 0.9,
        "gpu": false
      }
    }],
    "env": [{
      "name": "LunarLander-v2",
      "frame_op": "concat",
      "frame_op_len": 4,
      "max_t": null,
      "max_frame": 300000
    }],
    "body": {
      "product": "outer",
      "num": 1
    },
    "meta": {
      "distributed": false,
      "eval_frequency": 1000,
      "max_session": 4,
      "max_trial": 1,
    },
    "search": {
      "agent": [{
        "net": {
          "polyak_coef__grid_search": [0.8, 0.9, 0.95, 0.99]
        }
      }]
    }
  },
  "dqn_concat_replace_lunar": {
    "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": 0.08,
          "start_step": 0,
          "end_step": 12000
        },
        "gamma": 0.99,
        "training_batch_iter": 3,
        "training_iter": 4,
        "training_frequency": 4,
        "training_start_step": 32,
        "normalize_state": true,
      },
      "memory": {
        "name": "Replay",
        "batch_size": 32,
        "max_size": 100000,
        "use_cer": false
      },
      "net": {
        "type": "MLPNet",
        "hid_layers": [
          400,
          200
        ],
        "hid_layers_activation": "relu",
        "clip_grad_val": null,
        "loss_spec": {
          "name": "MSELoss"
        },
        "optim_spec": {
          "name": "Adam",
          "lr": 0.002
        },
        "lr_scheduler_spec": {
          "name": "MultiStepLR",
          "milestones": [
            80033
          ],
          "gamma": 0.1
        },
        "update_type": "replace",
        "update_frequency": 100,
        "gpu": false
      }
    }],
    "env": [{
      "name": "LunarLander-v2",
      "frame_op": "concat",
      "frame_op_len": 4,
      "max_t": null,
      "max_frame": 300000
    }],
    "body": {
      "product": "outer",
      "num": 1
    },
    "meta": {
      "distributed": false,
      "eval_frequency": 1000,
      "max_session": 4,
      "max_trial": 1,
    },
    "search": {
      "agent": [{
        "net": {
          "update_frequency__grid_search": [10, 100, 250, 500, 1000, 2000]
        }
      }]
    }
  },
  "vanilla_dqn_concat_lunar": {
    "agent": [{
      "name": "VanillaDQN",
      "algorithm": {
        "name": "VanillaDQN",
        "action_pdtype": "Argmax",
        "action_policy": "epsilon_greedy",
        "explore_var_spec": {
          "name": "linear_decay",
          "start_val": 1.0,
          "end_val": 0.08,
          "start_step": 0,
          "end_step": 80000
        },
        "gamma": 0.99,
        "training_batch_iter": 3,
        "training_iter": 4,
        "training_frequency": 4,
        "training_start_step": 32,
        "normalize_state": true,
      },
      "memory": {
        "name": "Replay",
        "batch_size": 32,
        "max_size": 100000,
        "use_cer": false
      },
      "net": {
        "type": "MLPNet",
        "hid_layers": [
          400,
          200
        ],
        "hid_layers_activation": "relu",
        "clip_grad_val": null,
        "loss_spec": {
          "name": "MSELoss"
        },
        "optim_spec": {
          "name": "Adam",
          "lr": 0.002
        },
        "lr_scheduler_spec": {
          "name": "MultiStepLR",
          "milestones": [
            80000
          ],
          "gamma": 0.1
        },
        "gpu": false
      }
    }],
    "env": [{
      "name": "LunarLander-v2",
      "frame_op": "concat",
      "frame_op_len": 4,
      "max_t": null,
      "max_frame": 300000
    }],
    "body": {
      "product": "outer",
      "num": 1
    },
    "meta": {
      "distributed": false,
      "eval_frequency": 1000,
      "max_session": 4,
      "max_trial": 1,
    },
    "search": {
      "agent": [{
        "algorithm": {
          "normalize_state__grid_search": [true, false],
        }
      }]
    }
  }
}