def determineRewards(BSs, Agents, actions, variables, t=-1, t_cutoff=-1):
x = determineWhichBSTransmitting(BSs, variables, t, t_cutoff);
Ntx = Counter(actions) #find number of UEs that chose each BS
rewards = [0]*len(Agents)
for i in range(0, len(Agents)):
if x[actions[i]] == 0: #BS it is connected to is not transmitting
continue
rewards[i] = environment.calculatecapacity(Agents[i], BSs[actions[i]], Ntx[actions[i]], variables, doFading=False)
return rewards
def act(self, BSs, variables,Agents=None, t = -1):
#calculate C to each BS
caps = [environment.calculatecapacity(self, BSs[0], 1, variables, doFading=False), environment.calculatecapacity(self, BSs[1], 1, variables, doFading=False)]
#multiply by K, 1-K
caps[0] = caps[0] * variables['K_coexistence']
caps[1] = caps[1] * (1-(variables['K_coexistence']))
#choose maximising
action = np.argmax(caps)
self.actions.append(action)
return action
def act(self, BSs, variables,Agents=None, t = -1):
Tpatience = math.floor(variables['T_cutoff']/2)
if t > Tpatience: #Above T_patience, it does basic learning does
p = 1-variables['p_explore'];
if random.random() < p:
avgOfEach = np.zeros(len(BSs))
for i in range(0,len(BSs)):
indices = [ind for ind, j in enumerate(self.actions) if j == i]
avgOfEach[i] = np.Inf if len(indices)==0 else sum([self.rewards[j] for j in indices])/(float(len(indices)))
action = np.argmax(avgOfEach)
else:
action = random.randint(0, len(BSs)-1)
else: #below a T_patience, it chooses the BS that would maximize K*C without other users
#calculate C to each BS
caps = [environment.calculatecapacity(self, BSs[0], 1, variables, doFading=False), environment.calculatecapacity(self, BSs[1], 1, variables, doFading=False)]
#multiply by K, 1-K
caps[0] = caps[0] * variables['K_coexistence']
caps[1] = caps[1] * (1-(variables['K_coexistence']))
#choose maximising
action = np.argmax(caps)
self.actions.append(action)
return action
else:
AgentRewards[i] = AgentRewards[i] + np.array(Agents[i].rewards)
AgentActions[i] = AgentActions[i] + np.array(Agents[i].actions)
foundcorre = False
if PLOTNEQ:
AllMixedStrategyRewards = []
AllMixedStrategyActions = []
# MixedStrategyRewards = np.zeros(variables['NumAgents'])
CorrelatedEquilRewards = np.zeros(variables["NumAgents"])
# MixedStrategyActions = np.zeros(variables['NumAgents'])
CorrelatedEquilActions = np.zeros(variables["NumAgents"])
C = np.zeros((2, 2))
for i in [0, 1]:
for j in [0, 1]:
C[i, j] = environment.calculatecapacity(Agents[i], BSs[j], 1, variables, doFading=False)
for strat in analysis_helper.findPossibleStrategies(C, variables["K_coexistence"]):
AllMixedStrategyRewards.append(
[
analysis_helper.calcUtilityFromStrategy(0, strat, variables["K_coexistence"], C),
analysis_helper.calcUtilityFromStrategy(1, strat, variables["K_coexistence"], C),
]
)
AllMixedStrategyActions.append([strat[0][1], strat[1][1]])
# bestmixed = analysis_helper.findBestMixedStrategy(C, variables["K_coexistence"])
# MixedStrategyRewards = MixedStrategyRewards + [analysis_helper.calcUtilityFromStrategy(0, bestmixed, variables["K_coexistence"], C), analysis_helper.calcUtilityFromStrategy(1, bestmixed, variables["K_coexistence"], C)]
# MixedStrategyActions = MixedStrategyActions + [bestmixed[0][1], bestmixed[1][1]]
foundcorre, corstrat, corrrewardsloc = analysis_helper.calc_correlated_equil(C, variables["K_coexistence"])
