File size: 6,561 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
173
174
175
176
177
178
179
180
181
182
from typing import Any, Union, List
import copy
import numpy as np
import gym
import competitive_rl

from ding.envs import BaseEnv, BaseEnvTimestep, BaseEnvInfo, update_shape
from ding.envs.common.env_element import EnvElement, EnvElementInfo
from ding.envs.common.common_function import affine_transform
from ding.torch_utils import to_ndarray, to_list
from .competitive_rl_env_wrapper import BuiltinOpponentWrapper, wrap_env
from ding.utils import ENV_REGISTRY

competitive_rl.register_competitive_envs()
"""
The observation spaces:
cPong-v0: Box(210, 160, 3)
cPongDouble-v0: Tuple(Box(210, 160, 3), Box(210, 160, 3))
cCarRacing-v0: Box(96, 96, 1)
cCarRacingDouble-v0: Box(96, 96, 1)

The action spaces:
cPong-v0: Discrete(3)
cPongDouble-v0: Tuple(Discrete(3), Discrete(3))
cCarRacing-v0: Box(2,)
cCarRacingDouble-v0: Dict(0:Box(2,), 1:Box(2,))

cPongTournament-v0
"""

COMPETITIVERL_INFO_DICT = {
    'cPongDouble-v0': BaseEnvInfo(
        agent_num=1,
        obs_space=EnvElementInfo(
            shape=(210, 160, 3),
            # shape=(4, 84, 84),
            value={
                'min': 0,
                'max': 255,
                'dtype': np.float32
            },
        ),
        act_space=EnvElementInfo(
            shape=(1, ),  # different with https://github.com/cuhkrlcourse/competitive-rl#usage
            value={
                'min': 0,
                'max': 3,
                'dtype': np.float32
            },
        ),
        rew_space=EnvElementInfo(
            shape=(1, ),
            value={
                'min': np.float32("-inf"),
                'max': np.float32("inf"),
                'dtype': np.float32
            },
        ),
        use_wrappers=None,
    ),
}


@ENV_REGISTRY.register('competitive_rl')
class CompetitiveRlEnv(BaseEnv):

    def __init__(self, cfg: dict) -> None:
        self._cfg = cfg
        self._env_id = self._cfg.env_id

        # opponent_type is used to control builtin opponent agent, which is useful in evaluator.
        is_evaluator = self._cfg.get("is_evaluator", False)
        opponent_type = None
        if is_evaluator:
            opponent_type = self._cfg.get("opponent_type", None)
        self._builtin_wrap = self._env_id == "cPongDouble-v0" and is_evaluator and opponent_type == "builtin"
        self._opponent = self._cfg.get('eval_opponent', 'RULE_BASED')

        self._init_flag = False

    def reset(self) -> np.ndarray:
        if not self._init_flag:
            self._env = self._make_env(only_info=False)
            self._init_flag = True
        if hasattr(self, '_seed') and hasattr(self, '_dynamic_seed') and self._dynamic_seed:
            np_seed = 100 * np.random.randint(1, 1000)
            self._env.seed(self._seed + np_seed)
        elif hasattr(self, '_seed'):
            self._env.seed(self._seed)
        obs = self._env.reset()
        obs = to_ndarray(obs)
        obs = self.process_obs(obs)  # process

        if self._builtin_wrap:
            self._eval_episode_return = np.array([0.])
        else:
            self._eval_episode_return = np.array([0., 0.])
        return obs

    def close(self) -> None:
        if self._init_flag:
            self._env.close()
        self._init_flag = False

    def seed(self, seed: int, dynamic_seed: bool = True) -> None:
        self._seed = seed
        self._dynamic_seed = dynamic_seed
        np.random.seed(self._seed)

    def step(self, action: Union[np.ndarray, list]) -> BaseEnvTimestep:
        action = to_ndarray(action)
        action = self.process_action(action)  # process

        obs, rew, done, info = self._env.step(action)

        if not isinstance(rew, tuple):
            rew = [rew]
        rew = np.array(rew)
        self._eval_episode_return += rew

        obs = to_ndarray(obs)
        obs = self.process_obs(obs)  # process

        if done:
            info['eval_episode_return'] = self._eval_episode_return

        return BaseEnvTimestep(obs, rew, done, info)

    def info(self) -> BaseEnvInfo:
        if self._env_id in COMPETITIVERL_INFO_DICT:
            info = copy.deepcopy(COMPETITIVERL_INFO_DICT[self._env_id])
            info.use_wrappers = self._make_env(only_info=True)
            obs_shape, act_shape, rew_shape = update_shape(
                info.obs_space.shape, info.act_space.shape, info.rew_space.shape, info.use_wrappers.split('\n')
            )
            info.obs_space.shape = obs_shape
            info.act_space.shape = act_shape
            info.rew_space.shape = rew_shape
            if not self._builtin_wrap:
                info.obs_space.shape = (2, ) + info.obs_space.shape
                info.act_space.shape = (2, )
                info.rew_space.shape = (2, )
            return info
        else:
            raise NotImplementedError('{} not found in COMPETITIVERL_INFO_DICT [{}]'\
                .format(self._env_id, COMPETITIVERL_INFO_DICT.keys()))

    def _make_env(self, only_info=False):
        return wrap_env(self._env_id, self._builtin_wrap, self._opponent, only_info=only_info)

    def __repr__(self) -> str:
        return "DI-engine Competitve RL Env({})".format(self._cfg.env_id)

    @staticmethod
    def create_collector_env_cfg(cfg: dict) -> List[dict]:
        collector_cfg = copy.deepcopy(cfg)
        collector_env_num = collector_cfg.pop('collector_env_num', 1)
        collector_cfg.is_evaluator = False
        return [collector_cfg for _ in range(collector_env_num)]

    @staticmethod
    def create_evaluator_env_cfg(cfg: dict) -> List[dict]:
        evaluator_cfg = copy.deepcopy(cfg)
        evaluator_env_num = evaluator_cfg.pop('evaluator_env_num', 1)
        evaluator_cfg.is_evaluator = True
        return [evaluator_cfg for _ in range(evaluator_env_num)]

    def process_action(self, action: np.ndarray) -> Union[tuple, dict, np.ndarray]:
        # If in double agent env, transfrom action passed in from outside to tuple or dict type.
        if self._env_id == "cPongDouble-v0" and not self._builtin_wrap:
            return (action[0].squeeze(), action[1].squeeze())
        elif self._env_id == "cCarRacingDouble-v0":
            return {0: action[0].squeeze(), 1: action[1].squeeze()}
        else:
            return action.squeeze()

    def process_obs(self, obs: Union[tuple, np.ndarray]) -> Union[tuple, np.ndarray]:
        # Copy observation for car racing double agent env, in case to be in alignment with pong double agent env.
        if self._env_id == "cCarRacingDouble-v0":
            obs = np.stack([obs, copy.deepcopy(obs)])
        return obs