from pybrain.tools.shortcuts import buildNetwork
from pybrain.rl.environments.cartpole import CartPoleEnvironment, CartPoleRenderer, BalanceTask
from pybrain.rl.agents.learning import LearningAgent
from pybrain.rl.experiments import EpisodicExperiment
from scipy import mean
import sys
episodes = 1
epilen = 200
if len(sys.argv) < 5:
sys.exit('please give 4 parameters. run: "python play_catpole.py <p1> <p2> <p3> <p4>"\n')
# create environment
env = CartPoleEnvironment()
env.setRenderer(CartPoleRenderer())
env.getRenderer().start()
env.delay = (episodes == 1)
# create task
task = BalanceTask(env, epilen)
# create controller network
net = buildNetwork(4, 1, bias=False)
# create agent and set parameters from command line
agent = LearningAgent(net, None)
agent.module._setParameters([float(sys.argv[1]), float(sys.argv[2]), float(sys.argv[3]), float(sys.argv[4])])
# create experiment
experiment = EpisodicExperiment(task, agent)
from pybrain.rl.experiments import EpisodicExperiment
from pybrain.rl.learners.valuebased import NFQ, ActionValueNetwork
from pybrain.rl.explorers import BoltzmannExplorer
from numpy import array, arange, meshgrid, pi, zeros, mean
from matplotlib import pyplot as plt
# switch this to True if you want to see the cart balancing the pole (slower)
render = False
#render = True
plt.ion()
env = CartPoleEnvironment()
if render:
renderer = CartPoleRenderer()
env.setRenderer(renderer)
renderer.start()
module = ActionValueNetwork(4, 3)
task = DiscreteBalanceTask(env, 100)
learner = NFQ()
learner.explorer.epsilon = 0.4
agent = LearningAgent(module, learner)
testagent = LearningAgent(module, None)
experiment = EpisodicExperiment(task, agent)
def plotPerformance(values, fig):
