q-iqn/variant.json

{
  "alg_class": {
    "$class": "rlkit.torch.torch_rl_algorithm.OfflineTorchBatchRLAlgorithm"
  },
  "algorithm": "sarsa-iqn",
  "algorithm_kwargs": {
    "batch_size": 256,
    "max_path_length": 0,
    "num_epochs": 0,
    "num_eval_steps_per_epoch": 0,
    "num_trains_per_train_loop": 1000,
    "start_epoch": -400
  },
  "d4rl": true,
  "env_id": "halfcheetah-medium-expert-v2",
  "normalize_env": true,
  "path_loader_kwargs": {},
  "pipeline": {
    "$class": "rlkit.launchers.pipeline.<Pipeline SarsaIQNPipeline>:\noffline_init,\ncreate_eval_env,\ncreate_dataset_next_actions,\noptionally_normalize_dataset,\ncreate_q_iqn,\ncreate_policy,\ncreate_trainer,\ncreate_eval_policy,\ncreate_eval_path_collector,\ncreate_replay_buffer,\ncreate_algorithm,\nload_demos,\ntrain"
  },
  "policy_class": {
    "$class": "rlkit.policies.gaussian_policy.TanhGaussianPolicy"
  },
  "policy_kwargs": {
    "hidden_sizes": [
      1024,
      1024
    ]
  },
  "qf_class": {
    "$class": "rlkit.torch.networks.mlp.QuantileMlp"
  },
  "qf_kwargs": {
    "embedding_size": 64,
    "hidden_sizes": [
      256,
      256,
      256
    ],
    "num_quantiles": 8
  },
  "replay_buffer_class": {
    "$class": "rlkit.data_management.env_replay_buffer.EnvReplayBufferNextAction"
  },
  "replay_buffer_size": 2000000,
  "rollout_fn": {
    "$function": "rlkit.samplers.rollout_functions.rollout"
  },
  "seed": 0,
  "snapshot_gap": 50,
  "snapshot_mode": "gap_and_last",
  "trainer_class": {
    "$class": "rlkit.torch.torch_rl_algorithm.TorchTrainer"
  },
  "trainer_cls": {
    "$class": "rlkit.torch.algorithms.sarsa_iqn.SarsaIQNTrainer"
  },
  "trainer_kwargs": {
    "discount": 0.99,
    "num_quantiles": 8,
    "policy_lr": 0.0003,
    "qf_lr": 0.0003,
    "reward_scale": 1,
    "soft_target_tau": 0.005,
    "target_update_period": 1
  },
  "version": "normalize-env"
}