-
Notifications
You must be signed in to change notification settings - Fork 1
/
Initial_str_KEG.py
272 lines (254 loc) · 12.6 KB
/
Initial_str_KEG.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
272
# ONLY FOR KEG: it removes quadratic term in best reaction which impacts the running time of creating the optimization model
from Instances import *
# get optimization software
import gurobipy as grb
import numpy as np
#######################################################
## COMPUTE INITIAL SET OF STRATEGIES ##
## Generate initial strategies: monopoly strategies ##
#######################################################
# INPUT
# G : game class (see Instances.py)
# OUTPUT
# S = list of strategies
# U_p = individial profit for each player and each strategy in S
# Best_m = list of the players best reaction models
def InitialStrategies(G,opt_solver=1):
### for each player produce the optimal strategy if she was alone in the game ###
S = [[[]] for p in range(G.m())] # list of strategies
U_p = [[[]] for p in range(G.m())] # associated individual profit
# Profile is the profile of strategies
Profile = [np.array([0 for k in range(G.n_I()[p]+G.n_C()[p])]) for p in range(G.m())]
# Best reaction models
Best_m = []
for p in range(G.m()):
try:
S[p][0], U_p[p][0], Model_p = BestReactionGurobi(G.m(),G.n_I()[p],G.n_C()[p],G.n_constr()[p],G.c()[p],G.Q()[p],G.A()[p],G.b()[p],Profile,p,False)
except:
print("Player ", p+1, " has no feasible solution or the problem is unbounded")
Best_m.append(Model_p)
return S, U_p, Best_m
#######################################################################################################################
#### ALTERNATIVE INITIALIZATION: social optimum or potential #######################################################################
#######################################################################################################################
# social = 1 then social optimum, otherwise, potential
def InitialStrategiesII(G,opt_solver=1,social=1):
if social:
S, U_p = SocialOptimumGurobi(G.m(), G.n_I(), G.n_C(), G.n_constr(), G.c(), G.Q(), G.A(), G.b())
else:
S, U_p = PotentialNEGurobi(G.m(), G.n_I(), G.n_C(), G.n_constr(), G.c(), G.Q(), G.A(), G.b())
return S, U_p, CreateModels(G.m(), G.n_I(), G.n_C(), G.n_constr(), G.c(), G.Q(), G.A(), G.b())
#######################################################################################################################
#### ALTERNATIVE INITIALIZATION: pure NE for potential part
#### Potential ###########################################################
#######################################################################################################################
def PotentialNEGurobi(m, n_I, n_C, n_constr, c, Q, A, b):
m_Pot = grb.Model("PotentialNE")
m_Pot.setParam( 'OutputFlag', False )
m_Pot.setParam("Threads", 2)
# set objective function direction
m_Pot.ModelSense = -1 # maximize
m_Pot.update()
x = [np.array([m_Pot.addVar(vtype="B", name="x_"+str(p)+'_'+str(i)) for i in range(n_I[p])]+[m_Pot.addVar(lb=0, vtype="C", name="x_"+str(p)+'_'+str(i)) for i in range(n_I[p],n_I[p]+n_C[p])]) for p in range(m)]
m_Pot.update()
QuadPart = grb.QuadExpr(0)
for p in range(m):
for k in range(n_constr[p]):
m_Pot.addConstr(np.dot(x[p],A[p][k]),grb.GRB.LESS_EQUAL, b[p][k])
m_Pot.update()
if n_I[p]+n_C[p] ==1:
QuadPart = QuadPart+grb.QuadExpr(x[p][0]*c[p][0]-0.5*x[p][0]*Q[p][p]*x[p][0])
else:
QuadPart = QuadPart+grb.QuadExpr(grb.quicksum(0.5*np.dot(x[j],np.dot((Q[p][j]+Q[j][p].T),x[p].T)) for j in range(p))+np.dot(x[p],c[p])-0.5*(np.dot(x[p],np.dot(Q[p][p],x[p].T))))
m_Pot.setObjective(QuadPart)
m_Pot.update()
#m_Pot.write("apagar.lp")
m_Pot.optimize()
try:
S = [[[x[p][k].x for k in range(n_I[p]+n_C[p])]] for p in range(m)]
U_p = [[float(np.dot(c[p],S[p][0])-0.5*np.dot(S[p][0],np.dot(Q[p][p],S[p][0])))] for p in range(m)]
return S,U_p
except:
print("No feasible profile of strategies", m_Pot.status)
return None
#######################################################################################################################
#### ALTERNATIVE INITIALIZATION: pure NE for potential part ###########################################################
#### Social optimum #######################
#######################################################################################################################
def SocialOptimumGurobi(m, n_I, n_C, n_constr, c, Q, A, b):
m_SO = grb.Model("SocialOptimum")
m_SO.setParam( 'OutputFlag', False )
m_SO.setParam("Threads", 2)
# set objective function direction
m_SO.ModelSense = -1 # maximize
m_SO.update()
x = [np.array([m_SO.addVar(vtype="B", name="x_"+str(p)+'_'+str(i)) for i in range(n_I[p])]+[m_SO.addVar(lb=0, vtype="C", name="x_"+str(p)+'_'+str(i)) for i in range(n_I[p],n_I[p]+n_C[p])]) for p in range(m)]
m_SO.update()
QuadPart = grb.QuadExpr(0)
for p in range(m):
for k in range(n_constr[p]):
m_SO.addConstr(np.dot(x[p],A[p][k]),grb.GRB.LESS_EQUAL, b[p][k])
m_SO.update()
if n_I[p]+n_C[p] ==1:
QuadPart = QuadPart+grb.QuadExpr(x[p][0]*c[p][0]-0.5*x[p][0]*Q[p][p]*x[p][0])
else:
QuadPart = QuadPart+grb.QuadExpr(grb.quicksum(np.dot(x[j],np.dot(Q[p][j],x[p].T)) for j in range(m) if j !=p)+np.dot(x[p],c[p])-0.5*(np.dot(x[p],np.dot(Q[p][p],x[p].T))))
m_SO.setObjective(QuadPart)
m_SO.update()
#m_SO.write("apagar.lp")
m_SO.optimize()
try:
S = [[[x[p][k].x for k in range(n_I[p]+n_C[p])]] for p in range(m)]
U_p = [[float(np.dot(c[p],S[p][0])-0.5*np.dot(S[p][0],np.dot(Q[p][p],S[p][0])))] for p in range(m)]
return S,U_p
except:
print("No feasible profile of strategies", m_SO.status)
return None
######################################################
def CreateModels(m, n_I, n_C, n_constr, c, Q, A, b):
Profile = [np.array([0 for k in range(n_I[p]+n_C[p])]) for p in range(m)]
Best_m = []
for p in range(m):
_,_,Model_p = BestReactionGurobi(m,n_I[p],n_C[p],n_constr[p],c[p],Q[p],A[p],b[p],Profile,p,True)
Best_m.append(Model_p)
return Best_m
##################################################################################
############## RESTRICTED STRATEGY METHOD ###############################
############## ###############################
############# to compute (EXACT) Nash equilibria ###############################
##################################################################################
# Compute Best Reaction of player against the strategy 'Profile'
def BestReactionGurobi(m,n_I_p,n_C_p,n_constr_p,c_p,Q_p,A_p,b_p,Profile,p,create, m_p = None,CE_verify=False):
if CE_verify:
xk_Qkp = sum(Q_p[k] for k in range(m) if k!=p)
else:
xk_Qkp = sum(np.dot(Profile[k], Q_p[k]) for k in range(m) if k!=p) # np.array
if m_p == None:
# initiate model
m_p = grb.Model("MIQPG")
# no pritting of the output
m_p.setParam( 'OutputFlag', False )
m_p.setParam("Threads", 2)
#m_p.setParam('BarHomogeneous', 1)
#m_p.setParam('DualReductions',0)
# set objective function direction
m_p.ModelSense = -1 # maximize
m_p.update()
# binary variables
x = [] # decision vector
for i in range(n_I_p):
#x.append(m.addVar(vtype="B", obj = float(c_p[i]+xk_Qkp[i]), name="x"+str(i)))
x.append(m_p.addVar(vtype="B", name="x"+str(i)))
m_p.update()
for i in range(n_I_p,n_C_p+n_I_p):
#x.append(m.addVar(lb=0, vtype="C", obj = float(c_p[i]+xk_Qkp[i]), name="x"+str(i)))
x.append(m_p.addVar(lb=0, vtype="C", name="x"+str(i)))
m_p.update()
x = np.array(x)
# constraints
for k in range(n_constr_p):
#m_p.addConstr(np.array(x).np.dot(A_p[k]), grb.GRB.LESS_EQUAL, b_p[k])
#m_p.addConstr(x.np.dot(A_p[k]), grb.GRB.LESS_EQUAL, b_p[k])
m_p.addConstr(np.dot(A_p[k],x), grb.GRB.LESS_EQUAL, b_p[k])
m_p.update()
if n_I_p+n_C_p ==1:
#QuadPart = grb.QuadExpr(x[0]*c_p[0]+ xk_Qkp*x[0]-0.5*x[0]*Q_p[p]*x[0])
QuadPart = grb.QuadExpr(x[0]*c_p[0])
else:
#QuadPart = grb.QuadExpr(x.np.dot(c_p)-0.5*(x.np.dot(Q_p[p].np.dot(x.T)))+xk_Qkp.np.dot(x.T))
#QuadPart = grb.QuadExpr(x.np.dot(c_p)-0.5*(x.np.dot(Q_p[p].np.dot(x.T))))
QuadPart = grb.QuadExpr(np.dot(c_p,x))
m_p.setObjective(QuadPart)
m_p.update()
if n_I_p+n_C_p ==1 and type(xk_Qkp) is not np.ndarray:
if CE_verify and m_p!=None:
# when we use CE, we change objective function in the indepedent part
#QuadPart = grb.QuadExpr(x[0]*c_p[0]-0.5*x[0]*Q_p[p]*x[0])
x_tmp = m_p.getVars()
QuadPart = grb.QuadExpr(x_tmp[0]*c_p[0])
m_p.setObjective(QuadPart)
m_p.update()
m_p.setObjective(m_p.getObjective()+xk_Qkp*x_tmp)
m_p.update()
else:
m_p.setObjective(m_p.getObjective()+xk_Qkp*m_p.getVars()[0])
m_p.update()
else:
if CE_verify and m_p!=None:
x_tmp = np.array(m_p.getVars())
#QuadPart = grb.QuadExpr(np.dot(c_p,m_p.getVars())-0.5*(np.dot(np.dot(m_p.getVars().T,Q_p[p]),m_p.getVars())))
QuadPart = grb.QuadExpr(np.dot(c_p,x_tmp))
#QuadPart = grb.QuadExpr(np.dot(c_p,x)-0.5*(np.dot(np.dot(x.T,Q_p[p]),x)))
m_p.setObjective(QuadPart)
m_p.update()
m_p.setObjective(m_p.getObjective()+np.dot(x_tmp,xk_Qkp))
m_p.update()
else:
m_p.setObjective(m_p.getObjective()+np.dot(m_p.getVars(),xk_Qkp))
m_p.update()
#m_p.write("apagar.lp")
# create is always false
if create:
if n_I_p+n_C_p ==1 and type(xk_Qkp) is not np.ndarray:
if CE_verify:
# when we use CE, we change objective function in the indepedent part
#QuadPart = grb.QuadExpr(x[0]*c_p[0]-0.5*x[0]*Q_p[p]*x[0])
x_tmp = m_p.getVars()
QuadPart = grb.QuadExpr(x_tmp[0]*c_p[0])
# overrite objective
m_p.setObjective(QuadPart)
m_p.update()
m_p.setObjective(m_p.getObjective()-xk_Qkp*m_p.getVars()[0])
else:
if CE_verify:
#QuadPart = grb.QuadExpr(np.dot(c_p,x)-0.5*(np.dot(np.dot(x.T,Q_p[p]),x)))
x_tmp = np.array(m_p.getVars())
QuadPart = grb.QuadExpr(np.dot(c_p,x_tmp))
# overrite objective
m_p.setObjective(QuadPart)
m_p.update()
m_p.setObjective(m_p.getObjective()-np.dot(xk_Qkp,m_p.getVars()))
m_p.update()
return None,None,m_p
if not CE_verify:
# warm start
for j,aux_var in enumerate(m_p.getVars()):
aux_var.start = Profile[p][j]
m_p.update()
m_p.optimize()
try:
#return [x[i].x for i in range(n_I_p+n_C_p)],m_p.ObjVal, m_p
sol = [i.x for i in m_p.getVars()]
value = m_p.ObjVal
if n_I_p+n_C_p ==1 and type(xk_Qkp) is not np.ndarray:
# this is important for NE verification
if CE_verify:
m_p.setObjective(m_p.getObjective()-xk_Qkp*x_tmp)
else:
m_p.setObjective(m_p.getObjective()-xk_Qkp*m_p.getVars()[0])
else:
if CE_verify:
m_p.setObjective(m_p.getObjective()-np.dot(x_tmp,xk_Qkp))
else:
# this is important for NE verfication
m_p.setObjective(m_p.getObjective()-np.dot(xk_Qkp,m_p.getVars()))
m_p.update()
return sol, value, m_p
except:
print("Wow! The best reaction problem has no feasible solution. The status code is: ", m_p.status)
if __name__ == "__main__":
G = Game("KEG",2,80,2,50,3)
# Verify best response
S = [[[]] for p in range(G.m())] # list of strategies
U_p = [[[]] for p in range(G.m())] # associated individual profit
# Profile is the profile of strategies
Profile = [np.array([0 for k in range(G.n_I()[p]+G.n_C()[p])]) for p in range(G.m())]
# Best reaction models
Best_m = []
p=0
S[p][0], U_p[p][0], Model_p = BestReactionGurobi(G.m(),G.n_I()[p],G.n_C()[p],G.n_constr()[p],G.c()[p],G.Q()[p],G.A()[p],G.b()[p],Profile,p,False)
#S, U_p, Best_m = InitialStrategies(G,1)
# verify initial STRATEGIES
#S, U_p, Best_m = InitialStrategies(G)
# verify initial strategies II: uses potential part of the game
#S_II, U_p_II, Best_m_II = InitialStrategiesII(G,1,1)