/
BaseReward.py
241 lines (185 loc) · 9.42 KB
/
BaseReward.py
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
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
# Copyright (c) 2019-2020, RTE (https://www.rte-france.com)
# See AUTHORS.txt
# This Source Code Form is subject to the terms of the Mozilla Public License, version 2.0.
# If a copy of the Mozilla Public License, version 2.0 was not distributed with this file,
# you can obtain one at http://mozilla.org/MPL/2.0/.
# SPDX-License-Identifier: MPL-2.0
# This file is part of Grid2Op, Grid2Op a testbed platform to model sequential decision making in power systems.
from abc import ABC, abstractmethod
from grid2op.dtypes import dt_float
class BaseReward(ABC):
"""
Base class from which all rewards used in the Grid2Op framework should derived.
In reinforcement learning, a reward is a signal send by the :class:`grid2op.Environment.Environment` to the
:class:`grid2op.BaseAgent` indicating how well this agent performs.
One of the goal of Reinforcement Learning is to maximize the (discounted) sum of (expected) rewards over time.
You can create all rewards you want in grid2op. The only requirement is that all rewards should inherit this
BaseReward.
Attributes
----------
reward_min: ``float``
The minimum reward an :class:`grid2op.BaseAgent` can get performing the worst possible
:class:`grid2op.Action.BaseAction` in
the worst possible scenario.
reward_max: ``float``
The maximum reward an :class:`grid2op.Agent.BaseAgent` can get performing the best possible
:class:`grid2op.Action.BaseAction` in
the best possible scenario.
Examples
---------
If you want the environment to compute a reward that is the sum of the flow (this is not a good reward, but
we use it as an example on how to do it) you can achieve it with:
.. code-block:
import grid2op
from grid2op.Reward import BaseReward
# first you create your reward
class SumOfFlowReward(BaseReward):
def __init__(self):
BaseReward.__init__(self)
def initialize(self, env):
# this function is used to inform the class instance about the environment specification
# you can use `env.n_line` or `env.n_load` or `env.get_thermal_limit()` for example
# do not forget to initialize "reward_min" and "reward_max"
self.reward_min = 0.
self.reward_max = np.sum(env.get_thermal_limit)
# in this case the maximum reward is obtained when i compute the sum of the maximum flows
# on each powerline
def __call__(action, env, has_error, is_done, is_illegal, is_ambiguous):
# this method is called at the end of 'env.step' to compute the reward
# in our case we just want to sum the flow on each powerline because... why not...
if has_error:
# see the "Notes" paragraph for more information
res = self.reward_min
else:
res = np.sum(env.get_obs().a_or)
return res
# then you create your environment with it:
NAME_OF_THE_ENVIRONMENT = "rte_case14_realistic"
env = grid2op.make(NAME_OF_THE_ENVIRONMENT,reward_class=SumOfFlowReward)
# and do a step with a "do nothing" action
obs = env.reset()
obs, reward, done, info = env.step(env.action_space())
assert np.sum(obs.a_or) == reward
# the above should be true
Notes
------
If the flag `has_error` is set to ``True`` this indicates there has been an error in the "env.step" function.
This might induce some undefined behaviour if using some method of the environment.
Please make sure to check whether or not this is the case when defining your reward.
This "new" behaviour has been introduce to "fix" the akward behavior spotted in
# https://github.com/rte-france/Grid2Op/issues/146
.. code-block:: python
def __call__(action, env, has_error, is_done, is_illegal, is_ambiguous):
if has_error:
# DO SOMETHING IN THIS CASE
res = self.reward_min
else:
# DO NOT USE `env.get_obs()` (nor any method of the environment `env.XXX` if the flag `has_error`
# is set to ``True``
# This might result in undefined behaviour
res = np.sum(env.get_obs().a_or)
return res
"""
def __init__(self):
"""
Initializes :attr:`BaseReward.reward_min` and :attr:`BaseReward.reward_max`
"""
self.reward_min = dt_float(0.)
self.reward_max = dt_float(0.)
def initialize(self, env):
"""
If :attr:`BaseReward.reward_min`, :attr:`BaseReward.reward_max` or other custom attributes require to have a
valid :class:`grid2op.Environment.Environment` to be initialized, this should be done in this method.
**NB** reward_min and reward_max are used by the environment to compute the maximum and minimum reward and
cast it in "reward_range" which is part of the openAI gym public interface. If you don't define them, some
piece of code might not work as expected.
Parameters
----------
env: :class:`grid2op.Environment.Environment`
An environment instance properly initialized.
"""
pass
def reset(self, env):
"""
This method is called each time `env` is reset.
It can be usefull, for example if the reward depends on the length of the current chronics.
It does nothing by default.
Parameters
----------
env: :class:`grid2op.Environment.Environment`
The current environment
.. danger::
This function should not modify self.reward_min nor self.reward_max !!!
It might cause really hard trouble for agent to learn if you do so.
"""
pass
@abstractmethod
def __call__(self, action, env, has_error, is_done, is_illegal, is_ambiguous):
"""
Method called to compute the reward.
Parameters
----------
action: :class:`grid2op.Action.Action`
BaseAction that has been submitted by the :class:`grid2op.BaseAgent`
env: :class:`grid2op.Environment.Environment`
An environment instance properly initialized.
has_error: ``bool``
Has there been an error, for example a :class:`grid2op.DivergingPowerFlow` be thrown when the action has
been implemented in the environment.
is_done: ``bool``
Is the episode over (either because the agent has reached the end, or because there has been a game over)
is_illegal: ``bool``
Has the action submitted by the BaseAgent raised an :class:`grid2op.Exceptions.IllegalAction` exception.
In this case it has been replaced by "do nohting" by the environment. **NB** an illegal action is NOT
an ambiguous action. See the description of the Action module: :ref:`Illegal-vs-Ambiguous` for more details.
is_ambiguous: ``bool``
Has the action submitted by the BaseAgent raised an :class:`grid2op.Exceptions.AmbiguousAction` exception.
In this case it has been replaced by "do nothing" by the environment. **NB** an illegal action is NOT
an ambiguous action. See the description of the Action module: :ref:`Illegal-vs-Ambiguous` for more details.
Returns
-------
res: ``float``
The reward associated to the input parameters.
Notes
------
All the flags can be used to know on which type of situation the reward is computed.
For example, if `has_error` is ``True`` it means there was an error during the computation of the powerflow.
this means there is a "game_over", so ``is_done`` is ``True`` in this case.
But, if there is ``is_done`` equal to ``True`` but ``has_error`` equal to ``False`` this means that the episode
is over without any error. In other word, your agent sucessfully managed all the scenario and to get to the
end of the episode.
"""
pass
def get_range(self):
"""
Shorthand to retrieve both the minimum and maximum possible rewards in one command.
It is not recommended to override this function.
Returns
-------
reward_min: ``float``
The minimum reward, see :attr:`BaseReward.reward_min`
reward_max: ``float``
The maximum reward, see :attr:`BaseReward.reward_max`
"""
return self.reward_min, self.reward_max
def set_range(self, reward_min, reward_max):
"""
Setter function for the :attr:`BaseReward.reward_min` and :attr:`BaseReward.reward_max`.
It is not recommended to override this function
Parameters
-------
reward_min: ``float``
The minimum reward, see :attr:`BaseReward.reward_min`
reward_max: ``float``
The maximum reward, see :attr:`BaseReward.reward_max`
"""
self.reward_min = reward_min
self.reward_max = reward_max
def __iter__(self):
"""
Implements python iterable to get a dict summary using `summary = dict(reward_instance)`
Can be overloaded by subclass, default implementation gives name, reward_min, reward_max
"""
yield ("name", self.__class__.__name__)
yield ("reward_min", float(self.reward_min))
yield ("reward_max", float(self.reward_max))