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 | from abc import ABC, abstractmethod
from collections.abc import Iterable
import copy
from typing import Optional
import numpy as np
from numpy.typing import DTypeLike
class ActionNoise(ABC):
"""The action noise base class."""
def __init__(self) -> None:
super().__init__()
def reset(self) -> None: # noqa: B027
"""Call end of episode reset for the noise."""
pass
@abstractmethod
def __call__(self) -> np.ndarray:
raise NotImplementedError()
class NormalActionNoise(ActionNoise):
"""A Gaussian action noise.
Args:
mean: Mean value of the noise
sigma: Scale of the noise (std here)
dtype: Type of the output noise
"""
def __init__(self, mean: np.ndarray, sigma: np.ndarray, dtype: DTypeLike = np.float32) -> None:
self._mu = mean
self._sigma = sigma
self._dtype = dtype
super().__init__()
def __call__(self) -> np.ndarray:
return np.random.normal(self._mu, self._sigma).astype(self._dtype)
def __repr__(self) -> str:
return f"NormalActionNoise(mu={self._mu}, sigma={self._sigma})"
class OrnsteinUhlenbeckActionNoise(ActionNoise):
"""An Ornstein Uhlenbeck action noise, this is designed to approximate Brownian motion with friction.
Based on http://math.stackexchange.com/questions/1287634/implementing-ornstein-uhlenbeck-in-matlab
Args:
mean: Mean of the noise
sigma: Scale of the noise
theta: Rate of mean reversion
dt: Timestep for the noise
initial_noise: Initial value for the noise output, (if None: 0)
dtype: Type of the output noise
"""
def __init__(
self,
mean: np.ndarray,
sigma: np.ndarray,
theta: float = 0.15,
dt: float = 1e-2,
initial_noise: np.ndarray | None = None,
dtype: DTypeLike = np.float32,
) -> None:
self._theta = theta
self._mu = mean
self._sigma = sigma
self._dt = dt
self._dtype = dtype
self.initial_noise = initial_noise
self.noise_prev = np.zeros_like(self._mu)
self.reset()
super().__init__()
def __call__(self) -> np.ndarray:
noise = (
self.noise_prev
+ self._theta * (self._mu - self.noise_prev) * self._dt
+ self._sigma * np.sqrt(self._dt) * np.random.normal(size=self._mu.shape)
)
self.noise_prev = noise
return noise.astype(self._dtype)
def reset(self) -> None:
"""Reset the Ornstein Uhlenbeck noise, to the initial position."""
self.noise_prev = self.initial_noise if self.initial_noise is not None else np.zeros_like(self._mu)
def __repr__(self) -> str:
return f"OrnsteinUhlenbeckActionNoise(mu={self._mu}, sigma={self._sigma})"
class VectorizedActionNoise(ActionNoise):
"""A Vectorized action noise for parallel environments.
Args:
base_noise: Noise generator to use
n_envs: Number of parallel environments
"""
def __init__(self, base_noise: ActionNoise, n_envs: int) -> None:
try:
self.n_envs = int(n_envs)
assert self.n_envs > 0
except (TypeError, AssertionError) as e:
raise ValueError(f"Expected n_envs={n_envs} to be positive integer greater than 0") from e
self.base_noise = base_noise
self.noises = [copy.deepcopy(self.base_noise) for _ in range(n_envs)]
def reset(self, indices: Iterable[int] | None = None) -> None:
"""Reset all the noise processes, or those listed in indices.
Args:
indices: The indices to reset. Default: None.
If the parameter is None, then all processes are reset to their initial position.
"""
if indices is None:
indices = range(len(self.noises))
for index in indices:
self.noises[index].reset()
def __repr__(self) -> str:
return f"VecNoise(BaseNoise={self.base_noise!r}), n_envs={len(self.noises)})"
def __call__(self) -> np.ndarray:
"""Generate and stack the action noise from each noise object."""
noise = np.stack([noise() for noise in self.noises])
return noise
@property
def base_noise(self) -> ActionNoise:
return self._base_noise
@base_noise.setter
def base_noise(self, base_noise: ActionNoise) -> None:
if base_noise is None:
raise ValueError("Expected base_noise to be an instance of ActionNoise, not None", ActionNoise)
if not isinstance(base_noise, ActionNoise):
raise TypeError("Expected base_noise to be an instance of type ActionNoise", ActionNoise)
self._base_noise = base_noise
@property
def noises(self) -> list[ActionNoise]:
return self._noises
@noises.setter
def noises(self, noises: list[ActionNoise]) -> None:
noises = list(noises) # raises TypeError if not iterable
assert len(noises) == self.n_envs, f"Expected a list of {self.n_envs} ActionNoises, found {len(noises)}."
different_types = [i for i, noise in enumerate(noises) if not isinstance(noise, type(self.base_noise))]
if len(different_types):
raise ValueError(
f"Noise instances at indices {different_types} don't match the type of base_noise",
type(self.base_noise),
)
self._noises = noises
for noise in noises:
noise.reset()
|