def getPossibleActions(self):
"""Returns the possible actions"""
#Cartesian product of all for all agents
allActions = []
for ag in range(0, self.numAg):
allActions.append(tuple(actions.all_agent_actions()))
#Returns all possible combined actions
listAct = self.cartesian(allActions).tolist()
ret = []
for ele in listAct:
ret.append(tuple(ele))
return ret
def initiateFromTL(self, state, action):
"""Reuses Q-values according to the paper description"""
sourceStates = self.translate_state(state)
#Defines the PITAM Mapping
okStates, totalSimilarityValue = PITAMUtil.pitam_mappings(
sourceStates, self.storedQTable)
#Finds best action according to PITAM
maxQ = -float('inf')
bestAct = None
for act in actions.all_agent_actions():
q = self.calculateQValue(okStates, act, totalSimilarityValue)
if maxQ < q:
maxQ = q
bestAct = act
#Initiates Q-entries
for act in actions.all_agent_actions():
self.qTable[(state, act)] = maxQ
#Only the best one receives the bias
self.qTable[(state, bestAct)] = self.bias + maxQ
return self.qTable[(state, action)]
def initiateFromTL(self, state, action):
"""Reuses Q-values according to the paper description"""
#Preventing errors regarding last state transition
if state == ('e', 'n', 'd'):
return 0
sourceStates = self.translate_state(state)
acts = actions.all_agent_actions()
voteActions = [0.0] * len(acts)
useBias = False
maxV = -float('inf')
maxAct = None
for sState in sourceStates:
maxV = -float('inf')
maxAct = None
for act in acts:
qV = self.storedQTable.get((sState, act), 0.0)
if qV > maxV:
maxV = qV
maxAct = act
if maxV > 0.0:
voteActions[maxAct] += maxV
useBias = True
for act in acts:
if (state, act) not in self.qTable:
self.qTable[(state, act)] = 0.0
#Bias transfer
if useBias:
maxAct = voteActions.index(max(voteActions))
self.qTable[(state, maxAct)] = self.bias
return self.qTable[(state, action)]
def getPossibleActions(self):
"""Returns the possible actions"""
return actions.all_agent_actions()
def select_action(self, state, agentIndex):
""" When this method is called, the agent executes an action. """
return random.choice(actions.all_agent_actions())