/
CombinedReward.py
71 lines (62 loc) · 2.63 KB
/
CombinedReward.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
# 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 grid2op.Reward.BaseReward import BaseReward
from grid2op.dtypes import dt_float
class CombinedReward(BaseReward):
"""
This class allows to combine multiple rewards, by summing them for example.
"""
def __init__(self):
BaseReward.__init__(self)
self.reward_min = dt_float(0.0)
self.reward_max = dt_float(0.0)
self.rewards = {}
def addReward(self, reward_name, reward_instance, reward_weight = 1.0):
self.rewards[reward_name] = {
"instance": reward_instance,
"weight": dt_float(reward_weight)
}
return True
def removeReward(self, reward_name):
if reward_name in self.rewards:
self.rewards.pop(reward_name)
return True
return False
def updateRewardWeight(self, reward_name, reward_weight):
if reward_name in self.rewards:
self.rewards[reward_name]["weight"] = reward_weight
return True
return False
def __iter__(self):
for k, v in super().__iter__():
yield (k, v)
for k, v in self.rewards.items():
r_dict = dict(v["instance"])
r_dict["weight"] = float(v["weight"])
yield (k, r_dict)
def initialize(self, env):
self.reward_min = dt_float(0.0)
self.reward_max = dt_float(0.0)
for key, reward in self.rewards.items():
reward_w = reward["weight"]
reward_instance = reward["instance"]
reward_instance.initialize(env)
self.reward_max += dt_float(reward_instance.reward_max * reward_w)
self.reward_min += dt_float(reward_instance.reward_min * reward_w)
def __call__(self, action, env, has_error, is_done, is_illegal, is_ambiguous):
res = dt_float(0.0)
# Loop over registered rewards
for key, reward in self.rewards.items():
r_instance = reward["instance"]
# Call individual reward
r = r_instance(action, env, has_error, is_done, is_illegal, is_ambiguous)
# Sum by weighted result
w = dt_float(reward["weight"])
res += dt_float(r) * w
# Return total sum
return res