hiddenUnits = 4
batch=1 #number of samples per learning step
prnts=1 #number of learning steps after results are printed
epis=10000/batch/prnts #number of roleouts
numbExp=10 #number of experiments
et = ExTools(batch, prnts) #tool for printing and plotting
env = None
for runs in range(numbExp):
# create environment
#Options: XML-Model, Bool(OpenGL), Bool(Realtime simu. while client is connected), ServerIP(default:localhost), Port(default:21560)
if env != None: env.closeSocket()
env = CCRLEnvironment()
# create task
task = CCRLGlasTask(env)
# create controller network
net = buildNetwork(len(task.getObservation()), hiddenUnits, env.actLen, outclass=TanhLayer)
# create agent with controller and learner (and its options)
agent = OptimizationAgent(net, PGPE(storeAllEvaluations = True))
et.agent = agent
# create the experiment
experiment = EpisodicExperiment(task, agent)
#Do the experiment
for updates in range(epis):
for i in range(prnts):
experiment.doEpisodes(batch)
et.printResults((agent.learner._allEvaluations)[-50:-1], runs, updates)
et.addExps()
et.showExps()
__author__ = 'Stubborn'
from pybrain.rl.environments.ode import CCRLEnvironment
from pybrain.rl.environments.ode.tasks import CCRLGlasTask
from pybrain.tools.shortcuts import buildNetwork
from pybrain.structure.modules.tanhlayer import TanhLayer
from pybrain.optimization import PGPE
from pybrain.rl.agents import OptimizationAgent
from pybrain.rl.experiments import EpisodicExperiment
environment = CCRLEnvironment()
task = CCRLGlasTask(environment)
net = buildNetwork(len(task.getObservation()), 4, environment.indim, outclass=TanhLayer)
agent = OptimizationAgent(net, PGPE())
experiment = EpisodicExperiment(task, agent)
for updates in range(20000):
experiment.doEpisodes(1)
hiddenUnits = 4
batch = 1 #number of samples per learning step
prnts = 1 #number of learning steps after results are printed
epis = 2000 / batch / prnts #number of roleouts
numbExp = 10 #number of experiments
et = ExTools(batch, prnts) #tool for printing and plotting
env = None
for runs in range(numbExp):
# create environment
#Options: XML-Model, Bool(OpenGL), Bool(Realtime simu. while client is connected), ServerIP(default:localhost), Port(default:21560)
if env != None: env.closeSocket()
env = CCRLEnvironment()
# create task
task = CCRLGlasTask(env)
# create controller network
net = buildNetwork(len(task.getObservation()),
hiddenUnits,
env.actLen,
outclass=TanhLayer) #, hiddenUnits
# create agent with controller and learner (and its options)
agent = OptimizationAgent(net, PGPE(storeAllEvaluations=True))
et.agent = agent
# create the experiment
experiment = EpisodicExperiment(task, agent)
#Do the experiment
for updates in range(epis):
for i in range(prnts):
experiment.doEpisodes(batch)
