-
-
Notifications
You must be signed in to change notification settings - Fork 3
/
Source_Code_MARL.py
271 lines (213 loc) · 10.2 KB
/
Source_Code_MARL.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
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
#importing dependencies
import numpy as np
import random
import pandas as pd
import matplotlib.pyplot as plt
plt.style.use('seaborn')
#loading dataFrames
LT_df = pd.read_csv('./metadata/LT_db.csv', index_col=0)
MT_df = pd.read_csv('./metadata/MT_db.csv', index_col=0)
ST_df = pd.read_csv('./metadata/ST_db.csv', index_col=0)
years_array=np.arange(int(LT_df.columns[0])-1,int(LT_df.columns[-1])+1)
global_trend_Real = [4.025, 4.074, 4.124, 4.152, 4.227, 4.224, 4.194, 4.173, 4.068, 4.002,4.011, 4.036, 4.071, 4.082, 4.05, 3.968, 4.038, 4.081, 4.088, 4.258, 4.414, 4.528, 4.636, 4.671, 4.762, 4.662, 4.835, 4.975, 5.005, 4.998, 4.981]
years = len(LT_df.columns)
#initializing all global variables
#initializing variable that counts the state periods
epochs = 32
number_of_agents = 3
#for graphical purposes; showing the action space in the first row of the Q-table
action_count = (" -0.2 -0.16 -0.12 -0.08 -0.04 0 0.04 0.08 0.12 0.16 0.2")
global_state = 4.97 #current CO2 emissions are at 4.97 metric tons per capita world-wide. Source: World Bank
#creating a list to store the global state for each epoch
global_state_per_epoch = []
#number of different actions that a single agent can take in a given state
size_of_action_space = 10
cost_of_action = 10 #defining cost to reduce CO2 emissions per metric ton
#Q-tables
Q_LT = np.zeros((years + 1, size_of_action_space))
Q_MT = np.zeros((years + 1, size_of_action_space))
Q_ST = np.zeros((years + 1, size_of_action_space))
#Q-tables for every epoch that stores all max values from
Q_LT_per_epoch = []
Q_MT_per_epoch = []
Q_ST_per_epoch = []
#defining the weight factors of immediate rewards
LT_reward_factor = 0.4
MT_reward_factor = 0.5
ST_reward_factor = 0.6
cumulative_reward = 0 #initializing cumulative reward, which is 0 to start with
#creating a list to store the cumulative reward for each epoch
cumulative_reward_per_epoch = []
#creating a list to store the immediate rewards for each epoch
immediate_rewards_per_epoch = []
LT_epsilon_min = 0.1 #defining minimal epsilon for LT
LT_epsilon_decay = 0.999 #defining decay rate of LT's epsilon
MT_epsilon_min = 0.06
MT_epsilon_decay = 0.995
ST_epsilon_min = 0.03
ST_epsilon_decay = 0.990
alpha = 0.1 #initializing the learning rate of the Q-values
alpha_min = 0.01 #initializing minimal learning rate after decay
alpha_decay = 0.999 #initializing decay of learning rate
gamma = 0.7 #<=>reward discount
LT_values = LT_df.values[:-1].T
Population_LT = LT_df.values[-1]
LT_ravel = LT_values.ravel()
LT_avg = sum(LT_ravel) / len(LT_ravel)
LT_action_space = np.round(
np.concatenate(
(np.linspace(min(LT_ravel), LT_avg, size_of_action_space // 2),
np.linspace(LT_avg, max(LT_ravel),
size_of_action_space - (size_of_action_space // 2)))), 3)
MT_values = MT_df.values[:-1].T
Population_MT = MT_df.values[-1]
MT_ravel = MT_values.ravel()
MT_avg = sum(MT_ravel) / len(MT_ravel)
MT_action_space = np.round(
np.concatenate(
(np.linspace(min(MT_ravel), MT_avg, size_of_action_space // 2),
np.linspace(MT_avg, max(MT_ravel),
size_of_action_space - (size_of_action_space // 2)))), 3)
ST_values = ST_df.values[:-1].T
Population_ST = ST_df.values[-1]
Population_array = Population_LT + Population_MT + Population_ST
ST_ravel = ST_values.ravel()
ST_avg = sum(ST_ravel) / len(ST_ravel)
ST_action_space = np.round(
np.concatenate(
(np.linspace(min(ST_ravel), ST_avg, size_of_action_space // 2),
np.linspace(ST_avg, max(ST_ravel),
size_of_action_space - (size_of_action_space // 2)))), 3)
#defining best actions from action space
Min_Q_LT = 0
Min_Q_MT = 0
Min_Q_ST = 0
for epoch in range(1, epochs + 1):
#initalize variables
global_state_of_co2_emission = 4.025 #CO2 emmision in 1984
cumulative_reward = 0 #intially 0
alpha = 0.1 #learning rate
#exploration vs explotation
LT_epsilon = 0.9
MT_epsilon = 0.8
ST_epsilon = 0.7
for year in range(0, years):
Min_Q_LT = np.argmin(Q_LT[year])
Min_Q_MT = np.argmin(Q_MT[year])
Min_Q_ST = np.argmin(Q_ST[year])
#LT
if np.random.rand() <= LT_epsilon:
LT_action = random.choice(LT_values[year])
else:
LT_action = LT_action_space[Min_Q_LT]
LT_immediate_reward = LT_action * (LT_reward_factor + cost_of_action)
Q_LT[year, abs(LT_action_space - LT_action).argmin()] = round(
(1 - alpha) * Q_LT[year, Min_Q_LT] + alpha *
(LT_immediate_reward + gamma * np.amin(Q_LT[year + 1, :])), 3)
#MT
if np.random.rand() <= MT_epsilon:
MT_action = random.choice(MT_values[year])
else:
MT_action = MT_action_space[Min_Q_MT]
MT_immediate_reward = MT_action * (MT_reward_factor + cost_of_action)
Q_MT[year, abs(MT_action_space - MT_action).argmin()] = round(
(1 - alpha) * Q_MT[year, Min_Q_MT] + alpha *
(MT_immediate_reward + gamma * np.amin(Q_MT[year + 1, :])), 3)
#ST
if np.random.rand() <= ST_epsilon:
ST_action = random.choice(ST_values[year])
else:
ST_action = ST_action_space[Min_Q_ST]
ST_immediate_reward = ST_action * (ST_reward_factor + cost_of_action)
Q_ST[year, abs(ST_action_space - ST_action).argmin()] = round(
(1 - alpha) * Q_ST[year, Min_Q_ST] + alpha *
(ST_immediate_reward + gamma * np.amin(Q_ST[year + 1, :])), 3)
cumulative_reward += LT_immediate_reward + MT_immediate_reward + ST_immediate_reward - cost_of_action * (
LT_action_space[np.argmin(Q_LT[year, :])] -
MT_action_space[np.argmin(Q_MT[year, :])] -
ST_action_space[np.argmin(Q_ST[year, :])])
alpha = alpha * alpha_decay if (alpha > alpha_min) else alpha
LT_epsilon = LT_epsilon * LT_epsilon_decay if (
LT_epsilon > LT_epsilon_min) else LT_epsilon_min
MT_epsilon = MT_epsilon * MT_epsilon_decay if (
MT_epsilon > MT_epsilon_min) else MT_epsilon_min
ST_epsilon = ST_epsilon * ST_epsilon_decay if (
ST_epsilon > ST_epsilon_min) else ST_epsilon_min
global_state_of_co2_emission += (
(LT_action * Population_LT[year]) +
(MT_action * Population_MT[year]) +
(ST_action * Population_ST[year])) / Population_array[year]
Q_LT_per_epoch.append(np.argmin(Q_LT, axis=1).tolist())
Q_MT_per_epoch.append(np.argmin(Q_MT, axis=1).tolist())
Q_ST_per_epoch.append(np.argmin(Q_ST, axis=1).tolist())
cumulative_reward_per_epoch.append(cumulative_reward)
immediate_rewards_per_epoch.append(
[LT_immediate_reward, MT_immediate_reward, ST_immediate_reward])
global_state_per_epoch.append(global_state_of_co2_emission)
immediate_rewards_per_epoch = np.array(immediate_rewards_per_epoch, copy=False)
#evaluating the trained model
print('Lowest Global State Achived During this Game: ',np.min(global_state_per_epoch))
print('\n')
print("Lowest Immediate Reward for LT: ",np.min(immediate_rewards_per_epoch[:, 0]))
print("Lowest Immediate Reward for MT: ",np.min(immediate_rewards_per_epoch[:, 1]))
print("Lowest Immediate Reward for ST: ",np.min(immediate_rewards_per_epoch[:, 2]))
print('Q_LT: \n')
print(pd.DataFrame(Q_LT).head())
print('Q_MT: \n')
print(pd.DataFrame(Q_MT).head())
print('Q_ST: \n')
print(pd.DataFrame(Q_ST).head())
#Synergistic Policy
Q_LT_Best = Q_LT_per_epoch[np.argmin(cumulative_reward_per_epoch)]
Q_MT_Best = Q_MT_per_epoch[np.argmin(cumulative_reward_per_epoch)]
Q_ST_Best = Q_ST_per_epoch[np.argmin(cumulative_reward_per_epoch)]
LT_Synergistic = [LT_action_space[i] for i in Q_LT_Best[1:]]
MT_Synergistic = [MT_action_space[i] for i in Q_MT_Best[1:]]
ST_Synergistic = [ST_action_space[i] for i in Q_ST_Best[1:]]
print("LT's Strategy to achieve Lowest Cumulative Reward:\n",LT_Synergistic)
print("\nMT's Strategy to achieve Lowest Cumulative Reward:\n",MT_Synergistic)
print("\nST's Strategy to achieve Lowest Cumulative Reward:\n",ST_Synergistic)
#Selfish Planning Policy
Q_LT_Immediate_Best = Q_LT_per_epoch[np.argmin(immediate_rewards_per_epoch[:, 0])]
Q_MT_Immediate_Best = Q_MT_per_epoch[np.argmin(immediate_rewards_per_epoch[:, 1])]
Q_ST_Immediate_Best = Q_ST_per_epoch[np.argmin(immediate_rewards_per_epoch[:, 2])]
LT_Selfish_Plan = [LT_action_space[i] for i in Q_LT_Immediate_Best[1:]]
MT_Selfish_Plan = [MT_action_space[i] for i in Q_MT_Immediate_Best[1:]]
ST_Selfish_Plan = [ST_action_space[i] for i in Q_ST_Immediate_Best[1:]]
print("LT's Strategy to achieve Lowest Immediate Reward:\n",LT_Selfish_Plan)
print("\nMT's Strategy to achieve Lowest Immediate Reward:\n",MT_Selfish_Plan)
print("\nST's Strategy to achieve Lowest Immediate Reward:\n",ST_Selfish_Plan)
#Greedy Policy
LT_Greedy = [LT_action_space[i] for i in np.argmin(Q_LT, axis=1)]
MT_Greedy = [MT_action_space[i] for i in np.argmin(Q_MT, axis=1)]
ST_Greedy = [ST_action_space[i] for i in np.argmin(Q_ST, axis=1)]
print("\nGreedy Policy of LT, based on LT's Final Q-Table:\n",LT_Greedy)
print("\nGreedy Policy of MT, based on MT's Final Q-Table:\n",MT_Greedy)
print("\nGreedy Policy of ST, based on ST's Final Q-Table:\n",ST_Greedy)
def get_trend(Q_LT,Q_MT,Q_ST,):
global_state_of_co2_emission = 4.025
global_trend = [4.025]
for year in range(0, years):
#actions based on Policies
LT_action = Q_LT[year]
MT_action = Q_MT[year]
ST_action = Q_ST[year]
global_state_of_co2_emission += (
(LT_action * Population_LT[year]) + (MT_action * Population_MT[year]) +
(ST_action * Population_ST[year])) / Population_array[year]
global_trend.append(np.round(global_state_of_co2_emission,3))
return global_trend
#ALL Policies
plt.figure(figsize=(10,8))
plt.plot(years_array,global_trend_Greedy)
plt.scatter(years_array,global_trend_Greedy)
plt.plot(years_array,global_trend_Selfish)
plt.scatter(years_array,global_trend_Selfish)
plt.plot(years_array,global_trend_Synergistic)
plt.scatter(years_array,global_trend_Synergistic)
plt.plot(years_array,global_trend_Real)
plt.scatter(years_array,global_trend_Real)
plt.xlabel('Years')
plt.ylabel('CO2 Emission')
plt.legend(['Greedy','Selfish','Synergistic','Real'])
plt.show()