environment.conditions['actionNum'] = 2
environment.centerCart = False
experiment = Experiment(environment, agent)
# cut off last two state dimensions
indexer = IndexingAdapter([0, 1], None)
experiment.addAdapter(indexer)
# add normalization adapter
normalizer = NormalizingAdapter(normalizeStates=[(-0.8, 0.8), (-12., 12.)])
experiment.addAdapter(normalizer)
if options.play:
renderer = CartPoleRenderer()
environment.renderer = renderer
renderer.start()
print "10 evaluation trials:"
valdata = experiment.evaluateEpisodes(10, visualize=False)
mean_return = mean([sum(v.rewards) for v in valdata])
print "mean return", mean_return
raise SystemExit
# add e-greedy exploration
explorer = EpsilonGreedyExplorer(0.3, episodeCount=1000)
experiment.addAdapter(explorer)
if os.path.exists('cart_play.saved'):
os.remove('cart_play.saved')
# run experiment
for i in range(1000):
