-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathrun_games.py
More file actions
294 lines (220 loc) · 8.1 KB
/
Copy pathrun_games.py
File metadata and controls
294 lines (220 loc) · 8.1 KB
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
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
import numpy as np
from collections import
class game:
def __init__(self, n_agents, n_states, n_actions):
self.n_agents = n_agents
self.n_states = n_states
self.n_actions = n_actions
self.counter = 0
def step(self, A):
R = None
S = None
return R, S
def duopoly(a1, a2, n_actions=6):
p1 = a1 / n_actions
p2 = a2 / n_actions
if p1 < p2:
r1 = (1 - p1) * p1
r2 = 0
if p1 == p2:
r1 = 0.5 * (1 - p1)
r2 = r1
if p1 > p2:
r1 = 0
r2 = (1 - p2) * p2
R = np.array([r1, r2])
S = np.array([a2, a1])
return R, S
def run_duopoly(EPSILON=0):
# Q = initialize_q_table(QINIT, N_AGENTS, N_STATES, N_ACTIONS)
Q = np.random.random((N_AGENTS, N_STATES, N_ACTIONS))
ALPHA = np.random.random_sample(size=N_AGENTS)
if EPSILON == "UNIFORM":
EPSILON = np.random.random_sample(size=N_AGENTS) * mask
else:
EPSILON = EPSILON * np.ones(N_AGENTS) * mask
EPS_START = EPSILON
EPS_END = EPSILON
EPS_DECAY = N_ITER / 8
M = {}
ind = np.arange(N_AGENTS)
S = np.random.randint(N_STATES, size=N_AGENTS)
R = np.ones(N_AGENTS) * -2
A = np.random.randint(N_STATES, size=N_AGENTS)
action1 = A[0]
action2 = A[1]
elist = []
for t in range(N_ITER):
EPSILON = (EPS_END + (EPS_START - EPS_END) * math.exp(-1. * t / EPS_DECAY)) # if t < N_ITER/10 else 0
elist.append(EPSILON)
A = e_greedy_select_action(Q, S, EPSILON)
if t % 2 == 0:
action1 = A[0]
else:
action2 = A[1]
R, S = duopoly(a1=action1, a2=action2, n_actions=N_ACTIONS)
Q, sum_of_belief_updates = bellman_update_q_table(Q, S, A, R, ALPHA, GAMMA)
### SAVE PROGRESS DATA
M[t] = {"nA": np.bincount(A, minlength=3),
"R": R,
"Qmean": Q.mean(axis=1).mean(axis=0),
# "groups": count_groups(Q[ind, S, :], 0.1),
"Qvar": Q[ind, S, :].var(axis=0),
"nA": np.bincount(A, minlength=3),
# "T": travel_time_per_route,
"sum_of_belief_updates": sum_of_belief_updates,
# "alignment": alignment,
# "recommendation_alignment": recommendation_alignment,
# "action_alignment": action_alignment,
}
return M, elist
def prisoners_dilemma(a1, a2, r, s):
if a1 == 0 and a2 == 0:
r1 = r
r2 = r
elif a1 == 0 and a2 == 1:
r1 = -s
r2 = 1
elif a1 == 1 and a2 == 0:
r1 = 1
r2 = -s
elif a1 == 1 and a2 == 1:
r1 = 0
r2 = 0
state = a1 + a2
R = np.array([r1, r2])
S = np.array([state, state])
return R, S
def run_prisoners(EPSILON=0):
# Q = initialize_q_table(QINIT, N_AGENTS, N_STATES, N_ACTIONS)
Q = np.random.random((N_AGENTS, N_STATES, N_ACTIONS)) - 0.5
ALPHA = np.random.random_sample(size=N_AGENTS)
if EPSILON == "UNIFORM":
EPSILON = np.random.random_sample(size=N_AGENTS) * mask
else:
EPSILON = EPSILON * np.ones(N_AGENTS) * mask
EPS_START = EPSILON
EPS_END = EPSILON
EPS_DECAY = N_ITER / 8
M = {}
ind = np.arange(N_AGENTS)
S = np.random.randint(N_STATES, size=N_AGENTS)
R = np.ones(N_AGENTS) * -2
A = np.random.randint(N_STATES, size=N_AGENTS)
action1 = A[0]
action2 = A[1]
elist = []
for t in range(N_ITER):
EPSILON = (EPS_END + (EPS_START - EPS_END) * math.exp(-1. * t / EPS_DECAY)) # if t < N_ITER/10 else 0
elist.append(EPSILON)
A = e_greedy_select_action(Q, S, EPSILON)
action1 = A[0]
action2 = A[1]
R, S = prisoners_dilemma(a1=action1, a2=action2, r=0.5, s=0.5)
Q, sum_of_belief_updates = bellman_update_q_table(Q, S, A, R, ALPHA, GAMMA)
### SAVE PROGRESS DATA
M[t] = {"nA": np.bincount(A, minlength=3),
"R": R,
"Qmean": Q.mean(axis=1).mean(axis=0),
# "groups": count_groups(Q[ind, S, :], 0.1),
"Qvar": Q[ind, S, :].var(axis=0),
"nA": np.bincount(A, minlength=3),
# "T": travel_time_per_route,
"sum_of_belief_updates": sum_of_belief_updates,
# "alignment": alignment,
# "recommendation_alignment": recommendation_alignment,
# "action_alignment": action_alignment,
}
return M, elist
def two_route_game(A, balance_parameter):
n_players = len(A)
fraction_a = (A == 0).sum() / n_players
fraction_b = 1 - fraction_a
travel_time_a = fraction_a + balance_parameter
travel_time_b = fraction_b + (1 - balance_parameter)
T = [-travel_time_a, -travel_time_b]
R = np.array([T[a] for a in A])
return R, T
def run_two_route_game(EPSILON=0, BALANCE=1):
Q = initialize_q_table(QINIT, N_AGENTS, N_STATES, N_ACTIONS)
# Q = np.random.random((N_AGENTS, N_STATES, N_ACTIONS)) * 2
ALPHA = np.random.random_sample(size=N_AGENTS)
if EPSILON == "UNIFORM":
EPSILON = np.random.random_sample(size=N_AGENTS) * mask
else:
EPSILON = EPSILON * np.ones(N_AGENTS) * mask
EPS_START = EPSILON
EPS_END = EPSILON
EPS_DECAY = N_ITER / 8
M = {}
ind = np.arange(N_AGENTS)
S = np.random.randint(N_STATES, size=N_AGENTS)
R = np.ones(N_AGENTS) * -2
A = np.random.randint(N_STATES, size=N_AGENTS)
elist = []
for t in range(N_ITER):
EPSILON = (EPS_END + (EPS_START - EPS_END) * math.exp(-1. * t / EPS_DECAY)) # if t < N_ITER/10 else 0
elist.append(EPSILON)
A = e_greedy_select_action(Q, S, EPSILON)
R, T = two_route_game(A, balance_parameter=BALANCE)
Q, sum_of_belief_updates = bellman_update_q_table(Q, S, A, R, ALPHA, GAMMA)
### SAVE PROGRESS DATA
M[t] = {"nA": np.bincount(A, minlength=3),
"R": R,
"Qmean": Q.mean(axis=1).mean(axis=0),
# "groups": count_groups(Q[ind, S, :], 0.1),
"Qvar": Q[ind, S, :].var(axis=0),
# "T": travel_time_per_route,
"sum_of_belief_updates": sum_of_belief_updates,
# "alignment": alignment,
# "recommendation_alignment": recommendation_alignment,
# "action_alignment": action_alignment,
}
return M, elist
def pigou(A):
n_agents = len(A)
n_up = (A == 0).sum()
n_down = (A == 1).sum()
pct = n_down / n_agents
r_0 = 1
r_1 = pct
T = [-r_0, -r_1]
R = np.array([T[a] for a in A]) # -1 * np.vectorize(dict_map.get)(A)
return R, T
def run_pigou_game(EPSILON=0, BALANCE=1):
Q = initialize_q_table(QINIT, N_AGENTS, N_STATES, N_ACTIONS)
# Q = np.random.random((N_AGENTS, N_STATES, N_ACTIONS)) * 2
ALPHA = np.random.random_sample(size=N_AGENTS)
if EPSILON == "DECAYED":
EPS_START = 1
EPS_END = 0
EPS_DECAY = N_ITER / 8
else:
EPS_START = EPSILON
EPS_END = EPSILON
EPS_DECAY = N_ITER / 8
M = {}
ind = np.arange(N_AGENTS)
S = np.random.randint(N_STATES, size=N_AGENTS)
R = np.ones(N_AGENTS) * -2
A = np.random.randint(N_STATES, size=N_AGENTS)
elist = []
for t in range(N_ITER):
EPSILON = (EPS_END + (EPS_START - EPS_END) * math.exp(-1. * t / EPS_DECAY)) # if t < N_ITER/10 else 0
elist.append(EPSILON)
A = e_greedy_select_action(Q, S, EPSILON)
R, T = pigou(A)
Q, sum_of_belief_updates = bellman_update_q_table(Q, S, A, R, ALPHA, GAMMA)
### SAVE PROGRESS DATA
M[t] = {"nA": np.bincount(A, minlength=3),
"R": R,
"Qmean": Q.mean(axis=1).mean(axis=0),
# "groups": count_groups(Q[ind, S, :], 0.1),
"Qvar": Q[ind, S, :].var(axis=0),
# "T": travel_time_per_route,
"sum_of_belief_updates": sum_of_belief_updates,
# "alignment": alignment,
# "recommendation_alignment": recommendation_alignment,
# "action_alignment": action_alignment,
}
return M, elist