def reset(self):
LoggingAgentSG.reset(self)
self._temperature = self.init_temperature
self._expl_proportion = self.init_exploration
self.learner.reset()
self._oaro = None
self.newEpisode()
def integrateObservation(self, obs):
if self.learning and not self.learner.batchMode and self.lastobs is not None:
if self.learner.passNextAction:
self._oaro = [self.lastobs, self.lastaction, self.lastreward, obs]
else:
self.learner._updateWeights(self.lastobs, self.lastaction, self.lastreward, obs)
LoggingAgentSG.integrateObservation(self, obs)
def reset(self):
LoggingAgentSG.reset(self)
self._temperature = self.init_temperature
self._expl_proportion = self.init_exploration
self.learner.reset()
self._oaro = None
self.newEpisode()
def integrateObservation(self, obs):
if self.learning and not self.learner.batchMode and self.lastobs is not None:
if self.learner.passNextAction:
self._oaro = [self.lastobs, self.lastaction, self.lastreward, obs]
else:
self.learner._updateWeights(self.lastobs, self.lastaction, self.lastreward, obs)
LoggingAgentSG.integrateObservation(self, obs)
def __init__(self, module, num_features, num_actions, num_agents, index,
learner):
"""
:key module: the acting module
:key learner: the learner (optional) """
assert isinstance(
learner,
IndexableValueBasedLearner), "learner should be indexable."
self.module = module
self.learner = learner
LoggingAgentSG.__init__(self, num_features, num_actions, num_agents,
index)
# if learner is available, tell it the module and data
if self.learner is not None:
self.learner.module = self.module
self.learner.dataset = self.history
self.learning = True
self.agentProperties["requireOtherAgentsState"] = False
self.agentProperties["requireJointAction"] = False
self.agentProperties["requireJointReward"] = False
#parity check
for prop in self.learner.getProperty().keys():
if learner.getProperty()[prop]:
assert self.getProperty(
)[prop], "learners property should same to that of agents."
def getAction(self):
""" Activate the module with the last observation, add the exploration from
the explorer object and store the result as last action. """
LoggingAgentSG.getAction(self)
self.lastaction = self.module.activate(self.lastobs)
if self.learning:
self.lastaction = self.learner.explore(self.lastobs, self.lastaction)
return self.lastaction
def __init__(self, learner, num_actions, numAgents, index, **kwargs):
assert isinstance(learner, IndexableValueBasedLearner), "learner should be indexable."
self.learner = learner
LoggingAgentSG.__init__(self, np.ones(numAgents)*learner.num_features, num_actions, numAgents, index, **kwargs)
self.learner._behaviorPolicy = self._actionProbs
self.reset()
self.agentProperties["requireOtherAgentsState"]=False
self.agentProperties["requireJointAction"]=True
self.agentProperties["requireJointReward"]=True
for prop in self.learner.getProperty().keys():
if learner.getProperty()[prop]:
assert self.getProperty()[prop], "learners property should same to that of agents."
def __init__(self, learner, num_features, num_actions, num_agents, index, **kwargs):
self.learner = learner
LoggingAgentSG.__init__(self, np.ones(num_agents, dtype=np.int8)*num_features, num_actions, num_agents, index, **kwargs)
assert isinstance(learner, IndexableValueBasedLearner), "learner should be indexable."
self.learner._behaviorPolicy = self._actionProbs
self.reset()
self.agentProperties["requireOtherAgentsState"]=False
self.agentProperties["requireJointAction"]=True
self.agentProperties["requireJointReward"]=True
for prop in self.learner.getProperty().keys():
if learner.getProperty()[prop]:
assert self.getProperty()[prop], "learners property should same to that of agents."
def getAction(self):
""" Activate the module with the last observation, add the exploration from
the explorer object and store the result as last action. """
LoggingAgentSG.getAction(self)
self.lastaction = self.module.activate(self.lastobs)
if self.learning:
self.lastaction = self.learner.explore(self.lastobs,
self.lastaction)
return self.lastaction
def __init__(self, learner, num_features, num_actions, num_agents, index,
**kwargs):
assert isinstance(learner,
NFCEQ), "learner should be instance of NFCEQ."
self.learner = learner
LoggingAgentSG.__init__(self, num_features, num_actions, num_agents,
index, **kwargs)
# if learner is available, tell it the module and data
if self.learner is not None:
self.learner.dataset = self.history
self.learning = True
self.learner._behaviorPolicy = self._actionProbs
self.reset()
self.agentProperties["requireOtherAgentsState"] = False
self.agentProperties["requireJointAction"] = True
self.agentProperties["requireJointReward"] = True
for prop in self.learner.getProperty().keys():
if learner.getProperty()[prop]:
assert self.getProperty(
)[prop], "learners property should same to that of agents."
def __init__(self, module, num_features, num_actions, num_agents, index, learner):
"""
:key module: the acting module
:key learner: the learner (optional) """
LoggingAgentSG.__init__(self, num_features, num_actions, num_agents,index)
assert isinstance(learner, IndexableValueBasedLearner), "learner should be indexable."
self.module = module
self.learner = learner
# if learner is available, tell it the module and data
if self.learner is not None:
self.learner.module = self.module
self.learner.dataset = self.history
self.learning = True
self.agentProperties["requireOtherAgentsState"]=False
self.agentProperties["requireJointAction"]=False
self.agentProperties["requireJointReward"]=False
#parity check
for prop in self.learner.getProperty().keys():
if learner.getProperty()[prop]:
assert self.getProperty()[prop], "learners property should same to that of agents."
def reset(self):
""" Clear the history of the agent and resets the module and learner. """
LoggingAgentSG.reset(self)
self.module.reset()
if self.learning:
self.learner.reset()
def integrateObservation(self, obs):
LoggingAgentSG.integrateObservation(self, obs)
def reset(self):
""" Clear the history of the agent and resets the module and learner. """
LoggingAgentSG.reset(self)
self.module.reset()
if self.learning:
self.learner.reset()
