ding.bonus.sac

ding.bonus.sac

SACAgent

Overview

Class of agent for training, evaluation and deployment of Reinforcement learning algorithm Soft Actor-Critic(SAC). For more information about the system design of RL agent, please refer to https://di-engine-docs.readthedocs.io/en/latest/03_system/agent.html.

Interface: __init__, train, deploy, collect_data, batch_evaluate, best

supported_env_list = list(supported_env_cfg.keys()) class-attribute instance-attribute

Overview

List of supported envs.

Examples: >>> from ding.bonus.sac import SACAgent >>> print(SACAgent.supported_env_list)

best property

Overview

Load the best model from the checkpoint directory, which is by default in folder exp_name/ckpt/eval.pth.tar. The return value is the agent with the best model.

Returns: - (:obj:SACAgent): The agent with the best model. Examples: >>> agent = SACAgent(env_id='LunarLanderContinuous-v2') >>> agent.train() >>> agent = agent.best

.. note:: The best model is the model with the highest evaluation return. If this method is called, the current model will be replaced by the best model.

__init__(env_id=None, env=None, seed=0, exp_name=None, model=None, cfg=None, policy_state_dict=None)

Overview

Initialize agent for SAC algorithm.

Arguments: - env_id (:obj:str): The environment id, which is a registered environment name in gym or gymnasium. If env_id is not specified, env_id in cfg.env must be specified. If env_id is specified, env_id in cfg.env will be ignored. env_id should be one of the supported envs, which can be found in supported_env_list. - env (:obj:BaseEnv): The environment instance for training and evaluation. If env is not specified, env_id`` or ``cfg.env.env_id`` must be specified. ``env_id`` or ``cfg.env.env_id`` will be used to create environment instance. If ``env`` is specified, ``env_id`` and ``cfg.env.env_id`` will be ignored. - seed (:obj:int): The random seed, which is set before running the program. Default to 0. - exp_name (:obj:str): The name of this experiment, which will be used to create the folder to save log data. Default to None. If not specified, the folder name will be ``env_id``-``algorithm``. - model (:obj:torch.nn.Module): The model of SAC algorithm, which should be an instance of class :class:ding.model.ContinuousQAC. If not specified, a default model will be generated according to the configuration. - cfg (:obj:Union[EasyDict, dict]): The configuration of SAC algorithm, which is a dict. Default to None. If not specified, the default configuration will be used. The default configuration can be found in ``ding/config/example/SAC/gym_lunarlander_v2.py``. - policy_state_dict (:obj:str`): The path of policy state dict saved by PyTorch a in local file. If specified, the policy will be loaded from this file. Default to None.

.. note:: An RL Agent Instance can be initialized in two basic ways. For example, we have an environment with id LunarLanderContinuous-v2 registered in gym, and we want to train an agent with SAC algorithm with default configuration. Then we can initialize the agent in the following ways: >>> agent = SACAgent(env_id='LunarLanderContinuous-v2') or, if we want can specify the env_id in the configuration: >>> cfg = {'env': {'env_id': 'LunarLanderContinuous-v2'}, 'policy': ...... } >>> agent = SACAgent(cfg=cfg) There are also other arguments to specify the agent when initializing. For example, if we want to specify the environment instance: >>> env = CustomizedEnv('LunarLanderContinuous-v2') >>> agent = SACAgent(cfg=cfg, env=env) or, if we want to specify the model: >>> model = ContinuousQAC(**cfg.policy.model) >>> agent = SACAgent(cfg=cfg, model=model) or, if we want to reload the policy from a saved policy state dict: >>> agent = SACAgent(cfg=cfg, policy_state_dict='LunarLanderContinuous-v2.pth.tar') Make sure that the configuration is consistent with the saved policy state dict.

train(step=int(10000000.0), collector_env_num=None, evaluator_env_num=None, n_iter_save_ckpt=1000, context=None, debug=False, wandb_sweep=False)

Overview

Train the agent with SAC algorithm for step iterations with collector_env_num collector environments and evaluator_env_num evaluator environments. Information during training will be recorded and saved by wandb.

Arguments: - step (:obj:int): The total training environment steps of all collector environments. Default to 1e7. - collector_env_num (:obj:int): The collector environment number. Default to None. If not specified, it will be set according to the configuration. - evaluator_env_num (:obj:int): The evaluator environment number. Default to None. If not specified, it will be set according to the configuration. - n_iter_save_ckpt (:obj:int): The frequency of saving checkpoint every training iteration. Default to 1000. - context (:obj:str): The multi-process context of the environment manager. Default to None. It can be specified as spawn, fork or forkserver. - debug (:obj:bool): Whether to use debug mode in the environment manager. Default to False. If set True, base environment manager will be used for easy debugging. Otherwise, subprocess environment manager will be used. - wandb_sweep (:obj:bool): Whether to use wandb sweep, which is a hyper-parameter optimization process for seeking the best configurations. Default to False. If True, the wandb sweep id will be used as the experiment name. Returns: - (:obj:TrainingReturn): The training result, of which the attributions are: - wandb_url (:obj:str): The weight & biases (wandb) project url of the trainning experiment.

deploy(enable_save_replay=False, concatenate_all_replay=False, replay_save_path=None, seed=None, debug=False)

Overview

Deploy the agent with SAC algorithm by interacting with the environment, during which the replay video can be saved if enable_save_replay is True. The evaluation result will be returned.

Arguments: - enable_save_replay (:obj:bool): Whether to save the replay video. Default to False. - concatenate_all_replay (:obj:bool): Whether to concatenate all replay videos into one video. Default to False. If enable_save_replay is False, this argument will be ignored. If enable_save_replay is True and concatenate_all_replay is False, the replay video of each episode will be saved separately. - replay_save_path (:obj:str): The path to save the replay video. Default to None. If not specified, the video will be saved in exp_name/videos. - seed (:obj:Union[int, List]): The random seed, which is set before running the program. Default to None. If not specified, self.seed will be used. If seed is an integer, the agent will be deployed once. If seed is a list of integers, the agent will be deployed once for each seed in the list. - debug (:obj:bool): Whether to use debug mode in the environment manager. Default to False. If set True, base environment manager will be used for easy debugging. Otherwise, subprocess environment manager will be used. Returns: - (:obj:EvalReturn): The evaluation result, of which the attributions are: - eval_value (:obj:np.float32): The mean of evaluation return. - eval_value_std (:obj:np.float32): The standard deviation of evaluation return.

collect_data(env_num=8, save_data_path=None, n_sample=None, n_episode=None, context=None, debug=False)

Overview

Collect data with SAC algorithm for n_episode episodes with env_num collector environments. The collected data will be saved in save_data_path if specified, otherwise it will be saved in exp_name/demo_data.

Arguments: - env_num (:obj:int): The number of collector environments. Default to 8. - save_data_path (:obj:str): The path to save the collected data. Default to None. If not specified, the data will be saved in exp_name/demo_data. - n_sample (:obj:int): The number of samples to collect. Default to None. If not specified, n_episode must be specified. - n_episode (:obj:int): The number of episodes to collect. Default to None. If not specified, n_sample must be specified. - context (:obj:str): The multi-process context of the environment manager. Default to None. It can be specified as spawn, fork or forkserver. - debug (:obj:bool): Whether to use debug mode in the environment manager. Default to False. If set True, base environment manager will be used for easy debugging. Otherwise, subprocess environment manager will be used.

batch_evaluate(env_num=4, n_evaluator_episode=4, context=None, debug=False)

Overview

Evaluate the agent with SAC algorithm for n_evaluator_episode episodes with env_num evaluator environments. The evaluation result will be returned. The difference between methods batch_evaluate and deploy is that batch_evaluate will create multiple evaluator environments to evaluate the agent to get an average performance, while deploy will only create one evaluator environment to evaluate the agent and save the replay video.

Arguments: - env_num (:obj:int): The number of evaluator environments. Default to 4. - n_evaluator_episode (:obj:int): The number of episodes to evaluate. Default to 4. - context (:obj:str): The multi-process context of the environment manager. Default to None. It can be specified as spawn, fork or forkserver. - debug (:obj:bool): Whether to use debug mode in the environment manager. Default to False. If set True, base environment manager will be used for easy debugging. Otherwise, subprocess environment manager will be used. Returns: - (:obj:EvalReturn): The evaluation result, of which the attributions are: - eval_value (:obj:np.float32): The mean of evaluation return. - eval_value_std (:obj:np.float32): The standard deviation of evaluation return.

Full Source Code

../ding/bonus/sac.py

1from typing import Optional, Union, List 2from ditk import logging 3from easydict import EasyDict 4import os 5import numpy as np 6import torch 7import treetensor.torch as ttorch 8from ding.framework import task, OnlineRLContext 9from ding.framework.middleware import CkptSaver, \ 10 wandb_online_logger, offline_data_saver, termination_checker, interaction_evaluator, StepCollector, data_pusher, \ 11 OffPolicyLearner, final_ctx_saver 12from ding.envs import BaseEnv 13from ding.envs import setup_ding_env_manager 14from ding.policy import SACPolicy 15from ding.utils import set_pkg_seed 16from ding.utils import get_env_fps, render 17from ding.config import save_config_py, compile_config 18from ding.model import ContinuousQAC 19from ding.model import model_wrap 20from ding.data import DequeBuffer 21from ding.bonus.common import TrainingReturn, EvalReturn 22from ding.config.example.SAC import supported_env_cfg 23from ding.config.example.SAC import supported_env 24 25 26class SACAgent: 27 """ 28 Overview: 29 Class of agent for training, evaluation and deployment of Reinforcement learning algorithm \ 30 Soft Actor-Critic(SAC). 31 For more information about the system design of RL agent, please refer to \ 32 <https://di-engine-docs.readthedocs.io/en/latest/03_system/agent.html>. 33 Interface: 34 ``__init__``, ``train``, ``deploy``, ``collect_data``, ``batch_evaluate``, ``best`` 35 """ 36 supported_env_list = list(supported_env_cfg.keys()) 37 """ 38 Overview: 39 List of supported envs. 40 Examples: 41 >>> from ding.bonus.sac import SACAgent 42 >>> print(SACAgent.supported_env_list) 43 """ 44 45 def __init__( 46 self, 47 env_id: str = None, 48 env: BaseEnv = None, 49 seed: int = 0, 50 exp_name: str = None, 51 model: Optional[torch.nn.Module] = None, 52 cfg: Optional[Union[EasyDict, dict]] = None, 53 policy_state_dict: str = None, 54 ) -> None: 55 """ 56 Overview: 57 Initialize agent for SAC algorithm. 58 Arguments: 59 - env_id (:obj:`str`): The environment id, which is a registered environment name in gym or gymnasium. \ 60 If ``env_id`` is not specified, ``env_id`` in ``cfg.env`` must be specified. \ 61 If ``env_id`` is specified, ``env_id`` in ``cfg.env`` will be ignored. \ 62 ``env_id`` should be one of the supported envs, which can be found in ``supported_env_list``. 63 - env (:obj:`BaseEnv`): The environment instance for training and evaluation. \ 64 If ``env`` is not specified, `env_id`` or ``cfg.env.env_id`` must be specified. \ 65 ``env_id`` or ``cfg.env.env_id`` will be used to create environment instance. \ 66 If ``env`` is specified, ``env_id`` and ``cfg.env.env_id`` will be ignored. 67 - seed (:obj:`int`): The random seed, which is set before running the program. \ 68 Default to 0. 69 - exp_name (:obj:`str`): The name of this experiment, which will be used to create the folder to save \ 70 log data. Default to None. If not specified, the folder name will be ``env_id``-``algorithm``. 71 - model (:obj:`torch.nn.Module`): The model of SAC algorithm, which should be an instance of class \ 72 :class:`ding.model.ContinuousQAC`. \ 73 If not specified, a default model will be generated according to the configuration. 74 - cfg (:obj:`Union[EasyDict, dict]`): The configuration of SAC algorithm, which is a dict. \ 75 Default to None. If not specified, the default configuration will be used. \ 76 The default configuration can be found in ``ding/config/example/SAC/gym_lunarlander_v2.py``. 77 - policy_state_dict (:obj:`str`): The path of policy state dict saved by PyTorch a in local file. \ 78 If specified, the policy will be loaded from this file. Default to None. 79 80 .. note:: 81 An RL Agent Instance can be initialized in two basic ways. \ 82 For example, we have an environment with id ``LunarLanderContinuous-v2`` registered in gym, \ 83 and we want to train an agent with SAC algorithm with default configuration. \ 84 Then we can initialize the agent in the following ways: 85 >>> agent = SACAgent(env_id='LunarLanderContinuous-v2') 86 or, if we want can specify the env_id in the configuration: 87 >>> cfg = {'env': {'env_id': 'LunarLanderContinuous-v2'}, 'policy': ...... } 88 >>> agent = SACAgent(cfg=cfg) 89 There are also other arguments to specify the agent when initializing. 90 For example, if we want to specify the environment instance: 91 >>> env = CustomizedEnv('LunarLanderContinuous-v2') 92 >>> agent = SACAgent(cfg=cfg, env=env) 93 or, if we want to specify the model: 94 >>> model = ContinuousQAC(**cfg.policy.model) 95 >>> agent = SACAgent(cfg=cfg, model=model) 96 or, if we want to reload the policy from a saved policy state dict: 97 >>> agent = SACAgent(cfg=cfg, policy_state_dict='LunarLanderContinuous-v2.pth.tar') 98 Make sure that the configuration is consistent with the saved policy state dict. 99 """ 100 101 assert env_id is not None or cfg is not None, "Please specify env_id or cfg." 102 103 if cfg is not None and not isinstance(cfg, EasyDict): 104 cfg = EasyDict(cfg) 105 106 if env_id is not None: 107 assert env_id in SACAgent.supported_env_list, "Please use supported envs: {}".format( 108 SACAgent.supported_env_list 109 ) 110 if cfg is None: 111 cfg = supported_env_cfg[env_id] 112 else: 113 assert cfg.env.env_id == env_id, "env_id in cfg should be the same as env_id in args." 114 else: 115 assert hasattr(cfg.env, "env_id"), "Please specify env_id in cfg." 116 assert cfg.env.env_id in SACAgent.supported_env_list, "Please use supported envs: {}".format( 117 SACAgent.supported_env_list 118 ) 119 default_policy_config = EasyDict({"policy": SACPolicy.default_config()}) 120 default_policy_config.update(cfg) 121 cfg = default_policy_config 122 123 if exp_name is not None: 124 cfg.exp_name = exp_name 125 self.cfg = compile_config(cfg, policy=SACPolicy) 126 self.exp_name = self.cfg.exp_name 127 if env is None: 128 self.env = supported_env[cfg.env.env_id](cfg=cfg.env) 129 else: 130 assert isinstance(env, BaseEnv), "Please use BaseEnv as env data type." 131 self.env = env 132 133 logging.getLogger().setLevel(logging.INFO) 134 self.seed = seed 135 set_pkg_seed(self.seed, use_cuda=self.cfg.policy.cuda) 136 if not os.path.exists(self.exp_name): 137 os.makedirs(self.exp_name) 138 save_config_py(self.cfg, os.path.join(self.exp_name, 'policy_config.py')) 139 if model is None: 140 model = ContinuousQAC(**self.cfg.policy.model) 141 self.buffer_ = DequeBuffer(size=self.cfg.policy.other.replay_buffer.replay_buffer_size) 142 self.policy = SACPolicy(self.cfg.policy, model=model) 143 if policy_state_dict is not None: 144 self.policy.learn_mode.load_state_dict(policy_state_dict) 145 self.checkpoint_save_dir = os.path.join(self.exp_name, "ckpt") 146 147 def train( 148 self, 149 step: int = int(1e7), 150 collector_env_num: int = None, 151 evaluator_env_num: int = None, 152 n_iter_save_ckpt: int = 1000, 153 context: Optional[str] = None, 154 debug: bool = False, 155 wandb_sweep: bool = False, 156 ) -> TrainingReturn: 157 """ 158 Overview: 159 Train the agent with SAC algorithm for ``step`` iterations with ``collector_env_num`` collector \ 160 environments and ``evaluator_env_num`` evaluator environments. Information during training will be \ 161 recorded and saved by wandb. 162 Arguments: 163 - step (:obj:`int`): The total training environment steps of all collector environments. Default to 1e7. 164 - collector_env_num (:obj:`int`): The collector environment number. Default to None. \ 165 If not specified, it will be set according to the configuration. 166 - evaluator_env_num (:obj:`int`): The evaluator environment number. Default to None. \ 167 If not specified, it will be set according to the configuration. 168 - n_iter_save_ckpt (:obj:`int`): The frequency of saving checkpoint every training iteration. \ 169 Default to 1000. 170 - context (:obj:`str`): The multi-process context of the environment manager. Default to None. \ 171 It can be specified as ``spawn``, ``fork`` or ``forkserver``. 172 - debug (:obj:`bool`): Whether to use debug mode in the environment manager. Default to False. \ 173 If set True, base environment manager will be used for easy debugging. Otherwise, \ 174 subprocess environment manager will be used. 175 - wandb_sweep (:obj:`bool`): Whether to use wandb sweep, \ 176 which is a hyper-parameter optimization process for seeking the best configurations. \ 177 Default to False. If True, the wandb sweep id will be used as the experiment name. 178 Returns: 179 - (:obj:`TrainingReturn`): The training result, of which the attributions are: 180 - wandb_url (:obj:`str`): The weight & biases (wandb) project url of the trainning experiment. 181 """ 182 183 if debug: 184 logging.getLogger().setLevel(logging.DEBUG) 185 logging.debug(self.policy._model) 186 # define env and policy 187 collector_env_num = collector_env_num if collector_env_num else self.cfg.env.collector_env_num 188 evaluator_env_num = evaluator_env_num if evaluator_env_num else self.cfg.env.evaluator_env_num 189 collector_env = setup_ding_env_manager(self.env, collector_env_num, context, debug, 'collector') 190 evaluator_env = setup_ding_env_manager(self.env, evaluator_env_num, context, debug, 'evaluator') 191 192 with task.start(ctx=OnlineRLContext()): 193 task.use( 194 interaction_evaluator( 195 self.cfg, 196 self.policy.eval_mode, 197 evaluator_env, 198 render=self.cfg.policy.eval.render if hasattr(self.cfg.policy.eval, "render") else False 199 ) 200 ) 201 task.use(CkptSaver(policy=self.policy, save_dir=self.checkpoint_save_dir, train_freq=n_iter_save_ckpt)) 202 task.use( 203 StepCollector( 204 self.cfg, 205 self.policy.collect_mode, 206 collector_env, 207 random_collect_size=self.cfg.policy.random_collect_size 208 if hasattr(self.cfg.policy, 'random_collect_size') else 0, 209 ) 210 ) 211 task.use(data_pusher(self.cfg, self.buffer_)) 212 task.use(OffPolicyLearner(self.cfg, self.policy.learn_mode, self.buffer_)) 213 task.use( 214 wandb_online_logger( 215 metric_list=self.policy._monitor_vars_learn(), 216 model=self.policy._model, 217 anonymous=True, 218 project_name=self.exp_name, 219 wandb_sweep=wandb_sweep, 220 ) 221 ) 222 task.use(termination_checker(max_env_step=step)) 223 task.use(final_ctx_saver(name=self.exp_name)) 224 task.run() 225 226 return TrainingReturn(wandb_url=task.ctx.wandb_url) 227 228 def deploy( 229 self, 230 enable_save_replay: bool = False, 231 concatenate_all_replay: bool = False, 232 replay_save_path: str = None, 233 seed: Optional[Union[int, List]] = None, 234 debug: bool = False 235 ) -> EvalReturn: 236 """ 237 Overview: 238 Deploy the agent with SAC algorithm by interacting with the environment, during which the replay video \ 239 can be saved if ``enable_save_replay`` is True. The evaluation result will be returned. 240 Arguments: 241 - enable_save_replay (:obj:`bool`): Whether to save the replay video. Default to False. 242 - concatenate_all_replay (:obj:`bool`): Whether to concatenate all replay videos into one video. \ 243 Default to False. If ``enable_save_replay`` is False, this argument will be ignored. \ 244 If ``enable_save_replay`` is True and ``concatenate_all_replay`` is False, \ 245 the replay video of each episode will be saved separately. 246 - replay_save_path (:obj:`str`): The path to save the replay video. Default to None. \ 247 If not specified, the video will be saved in ``exp_name/videos``. 248 - seed (:obj:`Union[int, List]`): The random seed, which is set before running the program. \ 249 Default to None. If not specified, ``self.seed`` will be used. \ 250 If ``seed`` is an integer, the agent will be deployed once. \ 251 If ``seed`` is a list of integers, the agent will be deployed once for each seed in the list. 252 - debug (:obj:`bool`): Whether to use debug mode in the environment manager. Default to False. \ 253 If set True, base environment manager will be used for easy debugging. Otherwise, \ 254 subprocess environment manager will be used. 255 Returns: 256 - (:obj:`EvalReturn`): The evaluation result, of which the attributions are: 257 - eval_value (:obj:`np.float32`): The mean of evaluation return. 258 - eval_value_std (:obj:`np.float32`): The standard deviation of evaluation return. 259 """ 260 261 if debug: 262 logging.getLogger().setLevel(logging.DEBUG) 263 # define env and policy 264 env = self.env.clone(caller='evaluator') 265 266 if seed is not None and isinstance(seed, int): 267 seeds = [seed] 268 elif seed is not None and isinstance(seed, list): 269 seeds = seed 270 else: 271 seeds = [self.seed] 272 273 returns = [] 274 images = [] 275 if enable_save_replay: 276 replay_save_path = os.path.join(self.exp_name, 'videos') if replay_save_path is None else replay_save_path 277 env.enable_save_replay(replay_path=replay_save_path) 278 else: 279 logging.warning('No video would be generated during the deploy.') 280 if concatenate_all_replay: 281 logging.warning('concatenate_all_replay is set to False because enable_save_replay is False.') 282 concatenate_all_replay = False 283 284 def single_env_forward_wrapper(forward_fn, cuda=True): 285 286 forward_fn = model_wrap(forward_fn, wrapper_name='base').forward 287 288 def _forward(obs): 289 # unsqueeze means add batch dim, i.e. (O, ) -> (1, O) 290 obs = ttorch.as_tensor(obs).unsqueeze(0) 291 if cuda and torch.cuda.is_available(): 292 obs = obs.cuda() 293 (mu, sigma) = forward_fn(obs, mode='compute_actor')['logit'] 294 action = torch.tanh(mu).detach().cpu().numpy()[0] # deterministic_eval 295 return action 296 297 return _forward 298 299 forward_fn = single_env_forward_wrapper(self.policy._model, self.cfg.policy.cuda) 300 301 # reset first to make sure the env is in the initial state 302 # env will be reset again in the main loop 303 env.reset() 304 305 for seed in seeds: 306 env.seed(seed, dynamic_seed=False) 307 return_ = 0. 308 step = 0 309 obs = env.reset() 310 images.append(render(env)[None]) if concatenate_all_replay else None 311 while True: 312 action = forward_fn(obs) 313 obs, rew, done, info = env.step(action) 314 images.append(render(env)[None]) if concatenate_all_replay else None 315 return_ += rew 316 step += 1 317 if done: 318 break 319 logging.info(f'DQN deploy is finished, final episode return with {step} steps is: {return_}') 320 returns.append(return_) 321 322 env.close() 323 324 if concatenate_all_replay: 325 images = np.concatenate(images, axis=0) 326 import imageio 327 imageio.mimwrite(os.path.join(replay_save_path, 'deploy.mp4'), images, fps=get_env_fps(env)) 328 329 return EvalReturn(eval_value=np.mean(returns), eval_value_std=np.std(returns)) 330 331 def collect_data( 332 self, 333 env_num: int = 8, 334 save_data_path: Optional[str] = None, 335 n_sample: Optional[int] = None, 336 n_episode: Optional[int] = None, 337 context: Optional[str] = None, 338 debug: bool = False 339 ) -> None: 340 """ 341 Overview: 342 Collect data with SAC algorithm for ``n_episode`` episodes with ``env_num`` collector environments. \ 343 The collected data will be saved in ``save_data_path`` if specified, otherwise it will be saved in \ 344 ``exp_name/demo_data``. 345 Arguments: 346 - env_num (:obj:`int`): The number of collector environments. Default to 8. 347 - save_data_path (:obj:`str`): The path to save the collected data. Default to None. \ 348 If not specified, the data will be saved in ``exp_name/demo_data``. 349 - n_sample (:obj:`int`): The number of samples to collect. Default to None. \ 350 If not specified, ``n_episode`` must be specified. 351 - n_episode (:obj:`int`): The number of episodes to collect. Default to None. \ 352 If not specified, ``n_sample`` must be specified. 353 - context (:obj:`str`): The multi-process context of the environment manager. Default to None. \ 354 It can be specified as ``spawn``, ``fork`` or ``forkserver``. 355 - debug (:obj:`bool`): Whether to use debug mode in the environment manager. Default to False. \ 356 If set True, base environment manager will be used for easy debugging. Otherwise, \ 357 subprocess environment manager will be used. 358 """ 359 360 if debug: 361 logging.getLogger().setLevel(logging.DEBUG) 362 if n_episode is not None: 363 raise NotImplementedError 364 # define env and policy 365 env_num = env_num if env_num else self.cfg.env.collector_env_num 366 env = setup_ding_env_manager(self.env, env_num, context, debug, 'collector') 367 368 if save_data_path is None: 369 save_data_path = os.path.join(self.exp_name, 'demo_data') 370 371 # main execution task 372 with task.start(ctx=OnlineRLContext()): 373 task.use( 374 StepCollector( 375 self.cfg, self.policy.collect_mode, env, random_collect_size=self.cfg.policy.random_collect_size 376 ) 377 ) 378 task.use(offline_data_saver(save_data_path, data_type='hdf5')) 379 task.run(max_step=1) 380 logging.info( 381 f'SAC collecting is finished, more than {n_sample} samples are collected and saved in `{save_data_path}`' 382 ) 383 384 def batch_evaluate( 385 self, 386 env_num: int = 4, 387 n_evaluator_episode: int = 4, 388 context: Optional[str] = None, 389 debug: bool = False 390 ) -> EvalReturn: 391 """ 392 Overview: 393 Evaluate the agent with SAC algorithm for ``n_evaluator_episode`` episodes with ``env_num`` evaluator \ 394 environments. The evaluation result will be returned. 395 The difference between methods ``batch_evaluate`` and ``deploy`` is that ``batch_evaluate`` will create \ 396 multiple evaluator environments to evaluate the agent to get an average performance, while ``deploy`` \ 397 will only create one evaluator environment to evaluate the agent and save the replay video. 398 Arguments: 399 - env_num (:obj:`int`): The number of evaluator environments. Default to 4. 400 - n_evaluator_episode (:obj:`int`): The number of episodes to evaluate. Default to 4. 401 - context (:obj:`str`): The multi-process context of the environment manager. Default to None. \ 402 It can be specified as ``spawn``, ``fork`` or ``forkserver``. 403 - debug (:obj:`bool`): Whether to use debug mode in the environment manager. Default to False. \ 404 If set True, base environment manager will be used for easy debugging. Otherwise, \ 405 subprocess environment manager will be used. 406 Returns: 407 - (:obj:`EvalReturn`): The evaluation result, of which the attributions are: 408 - eval_value (:obj:`np.float32`): The mean of evaluation return. 409 - eval_value_std (:obj:`np.float32`): The standard deviation of evaluation return. 410 """ 411 412 if debug: 413 logging.getLogger().setLevel(logging.DEBUG) 414 # define env and policy 415 env_num = env_num if env_num else self.cfg.env.evaluator_env_num 416 env = setup_ding_env_manager(self.env, env_num, context, debug, 'evaluator') 417 418 # reset first to make sure the env is in the initial state 419 # env will be reset again in the main loop 420 env.launch() 421 env.reset() 422 423 evaluate_cfg = self.cfg 424 evaluate_cfg.env.n_evaluator_episode = n_evaluator_episode 425 426 # main execution task 427 with task.start(ctx=OnlineRLContext()): 428 task.use(interaction_evaluator(self.cfg, self.policy.eval_mode, env)) 429 task.run(max_step=1) 430 431 return EvalReturn(eval_value=task.ctx.eval_value, eval_value_std=task.ctx.eval_value_std) 432 433 @property 434 def best(self) -> 'SACAgent': 435 """ 436 Overview: 437 Load the best model from the checkpoint directory, \ 438 which is by default in folder ``exp_name/ckpt/eval.pth.tar``. \ 439 The return value is the agent with the best model. 440 Returns: 441 - (:obj:`SACAgent`): The agent with the best model. 442 Examples: 443 >>> agent = SACAgent(env_id='LunarLanderContinuous-v2') 444 >>> agent.train() 445 >>> agent = agent.best 446 447 .. note:: 448 The best model is the model with the highest evaluation return. If this method is called, the current \ 449 model will be replaced by the best model. 450 """ 451 452 best_model_file_path = os.path.join(self.checkpoint_save_dir, "eval.pth.tar") 453 # Load best model if it exists 454 if os.path.exists(best_model_file_path): 455 policy_state_dict = torch.load(best_model_file_path, map_location=torch.device("cpu")) 456 self.policy.learn_mode.load_state_dict(policy_state_dict) 457 return self