File size: 7,979 Bytes
079c32c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 |
from typing import Union, Optional, List, Any, Tuple
import os
import torch
from ditk import logging
from functools import partial
from tensorboardX import SummaryWriter
from copy import deepcopy
from ding.envs import get_vec_env_setting, create_env_manager
from ding.worker import BaseLearner, InteractionSerialEvaluator, BaseSerialCommander, create_buffer, \
create_serial_collector
from ding.config import read_config, compile_config
from ding.policy import create_policy
from ding.reward_model import create_reward_model
from ding.utils import set_pkg_seed
from .utils import random_collect
def serial_pipeline_ngu(
input_cfg: Union[str, Tuple[dict, dict]],
seed: int = 0,
env_setting: Optional[List[Any]] = None,
model: Optional[torch.nn.Module] = None,
max_train_iter: Optional[int] = int(1e10),
max_env_step: Optional[int] = int(1e10),
) -> 'Policy': # noqa
"""
Overview:
Serial pipeline entry for NGU. The corresponding paper is
`never give up: learning directed exploration strategies`.
Arguments:
- input_cfg (:obj:`Union[str, Tuple[dict, dict]]`): Config in dict type. \
``str`` type means config file path. \
``Tuple[dict, dict]`` type means [user_config, create_cfg].
- seed (:obj:`int`): Random seed.
- env_setting (:obj:`Optional[List[Any]]`): A list with 3 elements: \
``BaseEnv`` subclass, collector env config, and evaluator env config.
- model (:obj:`Optional[torch.nn.Module]`): Instance of torch.nn.Module.
- max_train_iter (:obj:`Optional[int]`): Maximum policy update iterations in training.
- max_env_step (:obj:`Optional[int]`): Maximum collected environment interaction steps.
Returns:
- policy (:obj:`Policy`): Converged policy.
"""
if isinstance(input_cfg, str):
cfg, create_cfg = read_config(input_cfg)
else:
cfg, create_cfg = deepcopy(input_cfg)
create_cfg.policy.type = create_cfg.policy.type + '_command'
env_fn = None if env_setting is None else env_setting[0]
cfg = compile_config(cfg, seed=seed, env=env_fn, auto=True, create_cfg=create_cfg, save_cfg=True)
# Create main components: env, policy
if env_setting is None:
env_fn, collector_env_cfg, evaluator_env_cfg = get_vec_env_setting(cfg.env)
else:
env_fn, collector_env_cfg, evaluator_env_cfg = env_setting
collector_env = create_env_manager(cfg.env.manager, [partial(env_fn, cfg=c) for c in collector_env_cfg])
evaluator_env = create_env_manager(cfg.env.manager, [partial(env_fn, cfg=c) for c in evaluator_env_cfg])
# if you want to save replay, please uncomment this line
# evaluator_env.enable_save_replay(cfg.env.replay_path)
collector_env.seed(cfg.seed)
evaluator_env.seed(cfg.seed, dynamic_seed=False)
set_pkg_seed(cfg.seed, use_cuda=cfg.policy.cuda)
policy = create_policy(cfg.policy, model=model, enable_field=['learn', 'collect', 'eval', 'command'])
# Create worker components: learner, collector, evaluator, replay buffer, commander.
tb_logger = SummaryWriter(os.path.join('./{}/log/'.format(cfg.exp_name), 'serial'))
learner = BaseLearner(cfg.policy.learn.learner, policy.learn_mode, tb_logger, exp_name=cfg.exp_name)
collector = create_serial_collector(
cfg.policy.collect.collector,
env=collector_env,
policy=policy.collect_mode,
tb_logger=tb_logger,
exp_name=cfg.exp_name
)
evaluator = InteractionSerialEvaluator(
cfg.policy.eval.evaluator, evaluator_env, policy.eval_mode, tb_logger, exp_name=cfg.exp_name
)
replay_buffer = create_buffer(cfg.policy.other.replay_buffer, tb_logger=tb_logger, exp_name=cfg.exp_name)
commander = BaseSerialCommander(
cfg.policy.other.commander, learner, collector, evaluator, replay_buffer, policy.command_mode
)
rnd_reward_model = create_reward_model(cfg.rnd_reward_model, policy.collect_mode.get_attribute('device'), tb_logger)
episodic_reward_model = create_reward_model(
cfg.episodic_reward_model, policy.collect_mode.get_attribute('device'), tb_logger
)
# ==========
# Main loop
# ==========
# Learner's before_run hook.
learner.call_hook('before_run')
# Accumulate plenty of data at the beginning of training.
if cfg.policy.get('random_collect_size', 0) > 0:
random_collect(cfg.policy, policy, collector, collector_env, commander, replay_buffer)
estimate_cnt = 0
iter_ = 0
while True:
"""some hyper-parameters used in NGU"""
# index_to_eps = {i: 0.4 ** (1 + 8 * i / (self._env_num - 1)) for i in range(self._env_num)}
# index_to_beta = {
# i: 0.3 * torch.sigmoid(torch.tensor(10 * (2 * i - (collector_env_num - 2)) / (collector_env_num - 2)))
# for i in range(collector_env_num)
# }
# index_to_gamma = {
# i: 1 - torch.exp(
# (
# (collector_env_num - 1 - i) * torch.log(torch.tensor(1 - 0.997)) +
# i * torch.log(torch.tensor(1 - 0.99))
# ) / (collector_env_num - 1)
# )
# for i in range(collector_env_num)
# }
iter_ += 1
# Evaluate policy performance
if evaluator.should_eval(learner.train_iter):
stop, reward = evaluator.eval(learner.save_checkpoint, learner.train_iter, collector.envstep)
if stop:
break
# Collect data by default config n_sample/n_episode
new_data = collector.collect(train_iter=learner.train_iter, policy_kwargs=None)
# collect data for reward_model training
rnd_reward_model.collect_data(new_data)
episodic_reward_model.collect_data(new_data)
replay_buffer.push(new_data, cur_collector_envstep=collector.envstep)
# update reward_model
rnd_reward_model.train()
if (iter_ + 1) % cfg.rnd_reward_model.clear_buffer_per_iters == 0:
rnd_reward_model.clear_data()
episodic_reward_model.train()
if (iter_ + 1) % cfg.episodic_reward_model.clear_buffer_per_iters == 0:
episodic_reward_model.clear_data()
# Learn policy from collected data
for i in range(cfg.policy.learn.update_per_collect):
# Learner will train ``update_per_collect`` times in one iteration.
train_data = replay_buffer.sample(learner.policy.get_attribute('batch_size'), learner.train_iter)
if train_data is None:
# It is possible that replay buffer's data count is too few to train ``update_per_collect`` times
logging.warning(
"Replay buffer's data can only train for {} steps. ".format(i) +
"You can modify data collect config, e.g. increasing n_sample, n_episode."
)
break
# calculate the inter-episodic and episodic intrinsic reward
rnd_reward = rnd_reward_model.estimate(train_data)
episodic_reward = episodic_reward_model.estimate(train_data)
# update train_data reward using the augmented reward
train_data_augmented, estimate_cnt = episodic_reward_model.fusion_reward(
train_data,
rnd_reward,
episodic_reward,
nstep=cfg.policy.nstep,
collector_env_num=cfg.policy.collect.env_num,
tb_logger=tb_logger,
estimate_cnt=estimate_cnt
)
learner.train(train_data_augmented, collector.envstep)
if learner.policy.get_attribute('priority'):
replay_buffer.update(learner.priority_info)
if collector.envstep >= max_env_step or learner.train_iter >= max_train_iter:
break
# Learner's after_run hook.
learner.call_hook('after_run')
return policy
|