gamma = 1 #best:1 solved 460
epsilon = 1 #best:1 solved 460
mineps = 0.01 #best:0.01 solved 460 0.01 188
epsilonRate = 12 #best:40 solved 460 12 188 35 523
x = []
yscores = []
yeps = []
yalpha = []
environment = gym.make('CartPole-v0')
agent = QLearnTabular.QLearnTabular(nStates, environment, alpha, gamma,
epsilon)
streak = 0
for i in range(maxIter):
t = runEpisode.play(environment, agent, False)
x.append(i + 1)
yscores.append(agent.score)
yalpha.append(agent.alpha)
yeps.append(agent.epsilon)
if i + 1 >= 100:
if sum(yscores[-100:]) / 100 >= 195:
print("Solved after {} episodes!".format(i + 1))
break
agent.reset()
#agent.alpha = max(alpha * (0.85 ** (i//alphaRate)), minAlpha)
agent.epsilon = max(min(1, 1 - math.log10((i + 1) / epsilonRate)), mineps)
agent.alpha = agent.epsilon #best: solved 452
if i % 20 == 0:
print("Episode Number {}".format(i))
def randomPlay(environment, nIter=500):
"play random to produce mean and standard deviation"
agent = randomAgent(environment)
for _ in range(nIter):
runEpisode.play(environment, agent, False, False)
return agent.calcMeanStd()