def __init__(self, randomSpwan=False):
start = {"grid": 1, "x": 2, "y": 2}
goal = {"grid": 3, "x": 4, "y": 0}
self.gridWorld = GridWorld.GridWorld(start=start, goal=goal)
if randomSpwan:
agent = gridAgent.Agent(self.gridWorld, None)
else:
agent = gridAgent.Agent(self.gridWorld, None)
self.gridWorld.pieceItTogether()
#self.plotgrid = plotgrid.plotReward("DunRun Reward Plot")
self.episodeList = plotgrid.plotReward("Episode Graph")
def run(self, iterations, episodes):
totalSuccess = 0
#agentArray= [[0,3,3],[0,2,3], [0,1,3], [1,4,3],[2,4,3],[random.randint(0,4), random.randint(0,4), 3],[random.randint(0,4), random.randint(0,4), 3],[random.randint(0,4), random.randint(0,4), 3],[random.randint(0,4), random.randint(0,4), 3]]
#for i in range(episodes - len(agentArray) +1): agentArray.append([random.randint(0,4),random.randint(0,4),random.randint(0,len(self.gridWorld.arrayOfGrids)-1)+1])
for iter in range(1, iterations+1):
agent = gridAgent.Agent(2,2,1,self.policy)
agent.playerStateSetUp(self.gridWorld.understandingState())
for ep in range(1,episodes+1):
steps = 0 #step variable
while True:
print("Iter: {}, Ep: {}, {}".format(iter, ep, agent.playerStatus()))
#print(self.policy.epsilon)
#self.gridWorld.printOut(agent.agentState(), True)
agent.results(self.gridWorld.agentMove(agent.agentState(), agent.move()))
#if self.plotter: self.plotter.LogResults(agent.reward)
if self.sarsaLearning or self.qLearning: agent.sarsaUpdate()
steps += 1
print(self.policy.epsilon)
print("Iter: {}, Ep: {}, {}".format(iter, ep, agent.playerStatus())) if debug else False#debug
#Check if goal state has been reached?
if self.gridWorld.finished: totalSuccess += 1
if self.gridWorld.finished or steps > 5000:
#Reset goal state
self.gridWorld.finished = False
if self.plotter: self.plotter.LogResults(steps,ep)
break
#At end of episode, update policy and grab it from agent to give to new agent
if self.mcLearning: agent.mcUpdate()
if self.plotter: self.plotter.plot()
'''
if ep % 75 == 0:
agent.policy.epsilon -= 0.05
if agent.policy.epsilon < 0:
agent.policy.epsilon = 0
'''
self.policy = agent.policyRetrieval()
agent = gridAgent.Agent(2,2,1,self.policy)
#agent = gridAgent.Agent(agentArray[ep][0],agentArray[ep][1],agentArray[ep][2] ,self.policy)
print("The total # of successful runs: {}".format(totalSuccess))
self.policy.printOut()
def run(self, iterations, episodes, steps, printInfo = False):
totalSuccess = 0
agentArray= []
for i in range(episodes - len(agentArray) +1): agentArray.append([random.randint(0,4),random.randint(0,4),random.randint(0,len(self.gridWorld.arrayOfGrids)-1)+1])
for iter in range(1, iterations+1):
iterationStart = time.time()
writeUp = ""
agent = gridAgent.Agent(2,3,3,self.policy)
agent.playerStateSetUp(self.gridWorld.understandingState())
for ep in range(1,episodes+1):
episodeStart = time.time()
episodeAverageTime = 0
while True:
if printInfo:
print("Iter: {}, Ep: {}, {}".format(iter, ep, agent.playerStatus()))
self.gridWorld.printOut(agent.agentState(), True)
agent.results(self.gridWorld.agentMove(agent.agentState(), agent.move()))
if self.sarsaLearning or self.qLearning: agent.sarsaUpdate()
print("Iter: {}, Ep: {}, {}".format(iter, ep, agent.playerStatus())) if debug else False#debug
#Check if goal state has been reached?
if self.gridWorld.finished: totalSuccess += 1
if self.gridWorld.finished or agent.moves> steps:
#Reset goal state
self.gridWorld.finished = False
break
#At end of episode, update policy and grab it from agent to give to new agent
episodeEnd = time.time()
episodeAverageTime = episodeAverageTime + 1/ep * ( (episodeEnd - episodeStart) - episodeAverageTime)
if self.mcLearning: agent.mcUpdate()
self.policy = agent.policyRetrieval()
writeUp += "Episode {} ".format(ep) + agent.agentInformation() + " Time Taken: {} \n".format(episodeEnd - episodeStart)
#agent = gridAgent.Agent(1,2,3,self.policy)
agent = gridAgent.Agent(agentArray[ep][0],agentArray[ep][1],agentArray[ep][2] ,self.policy)
iterationEnd = time.time()
print("Last Iterations Time is: {} \nAverage Episode time is: {} with a total # of successful runs: {}".format(iterationEnd - iterationStart, episodeAverageTime, totalSuccess))
self.policy.printOut()
f = open(os.path.dirname(os.path.realpath(__file__)) + "\Algorithm {} Epsilon={} Alpha={} Gamma={}.txt".format(self.policy.name, self.policy.epsilon, self.policy.alpha, self.policy.gamma),"w")
f.write(writeUp)
f.close()