def runAgent(self, moduleDict, numExperiences):
agent = moduleDict['qlearningAgents'].QLearningAgent(**self.opts)
states = list(
filter(lambda state: len(self.grid.getPossibleActions(state)) > 0,
self.grid.getStates()))
sorted(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 = states[int(randObj.random() * len(states))]
# startState = randObj.choice(states)
s = self.grid.getPossibleActions(startState)
ss = randObj.random()
action = s[int(ss * len(s))]
# action = randObj.choice(s)
# 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 action not in qValues.keys():
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)
