def runAgent(self, moduleDict, numExperiences):
agent = moduleDict['qlearningAgents'].QLearningAgent(**self.opts)
states = filter(lambda state : len(self.grid.getPossibleActions(state)) > 0, self.grid.getStates())
states.sort()
randObj = FixedRandom().random
# choose a random start state and a random possible action from that state
# get the next state and reward from the transition function
lastExperience = None
for i in range(numExperiences):
startState = randObj.choice(states)
action = randObj.choice(self.grid.getPossibleActions(startState))
(endState, reward) = self.env.getRandomNextState(startState, action, randObj=randObj)
lastExperience = (startState, action, endState, reward)
agent.update(*lastExperience)
actions = list(reduce(lambda a, b: set(a).union(b), [self.grid.getPossibleActions(state) for state in states]))
values = {}
qValues = {}
policy = {}
for state in states:
values[state] = agent.computeValueFromQValues(state)
policy[state] = agent.computeActionFromQValues(state)
possibleActions = self.grid.getPossibleActions(state)
for action in actions:
if not qValues.has_key(action):
qValues[action] = {}
if action in possibleActions:
qValues[action][state] = agent.getQValue(state, action)
else:
qValues[action][state] = None
valuesPretty = self.prettyValues(values)
policyPretty = self.prettyPolicy(policy)
qValuesPretty = {}
for action in actions:
qValuesPretty[action] = self.prettyValues(qValues[action])
return (valuesPretty, qValuesPretty, actions, policyPretty, lastExperience)
def runAgent(self, moduleDict, numExperiences):
agent = moduleDict['qlearningAgents'].QLearningAgent(**self.opts)
states = filter(lambda state : len(self.grid.getPossibleActions(state)) > 0, self.grid.getStates())
states.sort()
randObj = FixedRandom().random
# choose a random start state and a random possible action from that state
# get the next state and reward from the transition function
lastExperience = None
for i in range(numExperiences):
startState = randObj.choice(states)
action = randObj.choice(self.grid.getPossibleActions(startState))
(endState, reward) = self.env.getRandomNextState(startState, action, randObj=randObj)
lastExperience = (startState, action, endState, reward)
agent.update(*lastExperience)
actions = list(reduce(lambda a, b: set(a).union(b), [self.grid.getPossibleActions(state) for state in states]))
values = {}
qValues = {}
policy = {}
for state in states:
values[state] = agent.computeValueFromQValues(state)
policy[state] = agent.computeActionFromQValues(state)
possibleActions = self.grid.getPossibleActions(state)
for action in actions:
if not qValues.has_key(action):
qValues[action] = {}
if action in possibleActions:
qValues[action][state] = agent.getQValue(state, action)
else:
qValues[action][state] = None
valuesPretty = self.prettyValues(values)
policyPretty = self.prettyPolicy(policy)
qValuesPretty = {}
for action in actions:
qValuesPretty[action] = self.prettyValues(qValues[action])
return (valuesPretty, qValuesPretty, actions, policyPretty, lastExperience)
def runAgent(self, moduleDict, numExperiences):
agent = moduleDict['qlearningAgents'].ApproximateQAgent(extractor=self.extractor, **self.opts)
states = [state for state in self.grid.getStates() if len(self.grid.getPossibleActions(state)) > 0]
states.sort()
randObj = FixedRandom().random
# choose a random start state and a random possible action from that state
# get the next state and reward from the transition function
lastExperience = None
for i in range(numExperiences):
startState = randObj.choice(states)
action = randObj.choice(self.grid.getPossibleActions(startState))
(endState, reward) = self.env.getRandomNextState(startState, action, randObj=randObj)
lastExperience = (startState, action, endState, reward)
agent.update(*lastExperience)
actions = list(reduce(lambda a, b: set(a).union(b), [self.grid.getPossibleActions(state) for state in states]))
qValues = {}
weights = agent.getWeights()
for state in states:
possibleActions = self.grid.getPossibleActions(state)
for action in actions:
if action not in qValues:
qValues[action] = {}
if action in possibleActions:
qValues[action][state] = agent.getQValue(state, action)
else:
qValues[action][state] = None
qValuesPretty = {}
for action in actions:
qValuesPretty[action] = self.prettyValues(qValues[action])
return (qValuesPretty, weights, actions, lastExperience)
def runAgent(self, moduleDict, numExperiences):
agent = moduleDict["qlearningAgents"].ApproximateQAgent(extractor=self.extractor, **self.opts)
states = filter(lambda state: len(self.grid.getPossibleActions(state)) > 0, self.grid.getStates())
states.sort()
randObj = FixedRandom().random
# choose a random start state and a random possible action from that state
# get the next state and reward from the transition function
lastExperience = None
for i in range(numExperiences):
startState = randObj.choice(states)
action = randObj.choice(self.grid.getPossibleActions(startState))
(endState, reward) = self.env.getRandomNextState(startState, action, randObj=randObj)
lastExperience = (startState, action, endState, reward)
agent.update(*lastExperience)
actions = list(reduce(lambda a, b: set(a).union(b), [self.grid.getPossibleActions(state) for state in states]))
qValues = {}
weights = agent.getWeights()
for state in states:
possibleActions = self.grid.getPossibleActions(state)
for action in actions:
if not qValues.has_key(action):
qValues[action] = {}
if action in possibleActions:
qValues[action][state] = agent.getQValue(state, action)
else:
qValues[action][state] = None
qValuesPretty = {}
for action in actions:
qValuesPretty[action] = self.prettyValues(qValues[action])
return (qValuesPretty, weights, actions, lastExperience)
def runAgent(self, moduleDict):
agent = moduleDict['qlearningAgents'].QLearningAgent(**self.opts)
states = filter(lambda state : len(self.grid.getPossibleActions(state)) > 0, self.grid.getStates())
states.sort()
randObj = FixedRandom().random
# choose a random start state and a random possible action from that state
# get the next state and reward from the transition function
for i in range(self.numExperiences):
startState = randObj.choice(states)
action = randObj.choice(self.grid.getPossibleActions(startState))
(endState, reward) = self.env.getRandomNextState(startState, action, randObj=randObj)
agent.update(startState, action, endState, reward)
return agent
def runAgent(self, moduleDict):
agent = moduleDict['qlearningAgents'].QLearningAgent(**self.opts)
states = filter(lambda state : len(self.grid.getPossibleActions(state)) > 0, self.grid.getStates())
states.sort()
randObj = FixedRandom().random
# choose a random start state and a random possible action from that state
# get the next state and reward from the transition function
for i in range(self.numExperiences):
startState = randObj.choice(states)
action = randObj.choice(self.grid.getPossibleActions(startState))
(endState, reward) = self.env.getRandomNextState(startState, action, randObj=randObj)
agent.update(startState, action, endState, reward)
return agent
def run_agent(self, module_dict, num_experiences):
agent = module_dict['q_learning_agents'].QLearningAgent(**self.opts)
states = [
state for state in self.grid.get_states()
if len(self.grid.get_possible_actions(state)) > 0
]
states.sort()
rand_obj = FixedRandom().random
# choose a random start state and a random possible action from that state
# get the next state and reward from the transition function
last_experience = None
for i in range(num_experiences):
start_state = rand_obj.choice(states)
action = rand_obj.choice(
self.grid.get_possible_actions(start_state))
(end_state,
reward) = self.env.get_random_next_state(start_state,
action,
rand_obj=rand_obj)
last_experience = (start_state, action, end_state, reward)
agent.update(*last_experience)
actions = list(
reduce(lambda a, b: set(a).union(b),
[self.grid.get_possible_actions(state)
for state in states]))
values = {}
q_values = {}
policy = {}
for state in states:
values[state] = agent.compute_value_from_q_values(state)
policy[state] = agent.compute_action_from_q_values(state)
possible_actions = self.grid.get_possible_actions(state)
for action in actions:
if action not in q_values:
q_values[action] = {}
if action in possible_actions:
q_values[action][state] = agent.get_q_value(state, action)
else:
q_values[action][state] = None
values_pretty = self.pretty_values(values)
policy_pretty = self.pretty_policy(policy)
q_values_pretty = {}
for action in actions:
q_values_pretty[action] = self.pretty_values(q_values[action])
return (values_pretty, q_values_pretty, actions, policy_pretty,
last_experience)
def run_agent(self, module_dict):
agent = module_dict['q_learning_agents'].QLearningAgent(**self.opts)
states = [
state for state in self.grid.get_states()
if len(self.grid.get_possible_actions(state)) > 0
]
states.sort()
rand_obj = FixedRandom().random
# choose a random start state and a random possible action from that state
# get the next state and reward from the transition function
for i in range(self.num_experiences):
start_state = rand_obj.choice(states)
action = rand_obj.choice(
self.grid.get_possible_actions(start_state))
(end_state,
reward) = self.env.get_random_next_state(start_state,
action,
rand_obj=rand_obj)
agent.update(start_state, action, end_state, reward)
return agent
def run_agent(self, module_dict, num_experiences):
agent = module_dict['q_learning_agents'].ApproximateQAgent(
extractor=self.extractor, **self.opts)
states = filter(
lambda state: len(self.grid.get_possible_actions(state)) > 0,
self.grid.get_states())
states.sort()
rand_obj = FixedRandom().random
# choose a random start state and a random possible action from that state
# get the next state and reward from the transition function
last_experience = None
for i in range(num_experiences):
start_state = rand_obj.choice(states)
action = rand_obj.choice(
self.grid.get_possible_actions(start_state))
(end_state,
reward) = self.env.get_random_next_state(start_state,
action,
rand_obj=rand_obj)
last_experience = (start_state, action, end_state, reward)
agent.update(*last_experience)
actions = list(
reduce(lambda a, b: set(a).union(b),
[self.grid.get_possible_actions(state)
for state in states]))
q_values = {}
weights = agent.get_weights()
for state in states:
possible_actions = self.grid.get_possible_actions(state)
for action in actions:
if not q_values.has_key(action):
q_values[action] = {}
if action in possible_actions:
q_values[action][state] = agent.get_q_value(state, action)
else:
q_values[action][state] = None
q_values_pretty = {}
for action in actions:
q_values_pretty[action] = self.pretty_values(q_values[action])
return (q_values_pretty, weights, actions, last_experience)