def runExperiment(opt, visualize_steps, visualize_learning, visualize_performance, q):
# Experiment要在子进程中创建，不能直接传创建好的对象（会影响logger的正常工作）
exp = Experiment(**opt)
# 给logger加handler
# 子进程的log->MemoryHandler->OutputHandler-> queue <-ExpOutputDialog.receive->QTextEdit
# log通过queue在进程间传递，主线程通过thread接收queue中的新消息
from logging.handlers import MemoryHandler
handler = MemoryHandler(capacity=1024, flushLevel=logging.INFO, target=OutputHandler(q))
exp.logger.addHandler(handler)
exp.run(visualize_steps=visualize_steps, # should each learning step be shown?
visualize_learning=visualize_learning, # show policy / value function?
visualize_performance=visualize_performance) # show performance runs?
exp.plot()
def __init__(self, domain, representation, policy,steps=100000):
opt = {}
opt["domain"] = domain
# Agent
opt["agent"] = Q_Learning(representation=representation, policy=policy,
discount_factor=domain.discount_factor,
initial_learn_rate=0.1,
learn_rate_decay_mode="boyan", boyan_N0=100,
lambda_=0.)
opt["checks_per_policy"] = 10
opt["max_steps"] = steps
opt["num_policy_checks"] = 20
experiment = Experiment(**opt)
experiment.run()
self.policy = opt["agent"].policy
self.domain = domain
def runExperiment(opt, visualize_steps, visualize_learning,
visualize_performance, q):
# Experiment要在子进程中创建，不能直接传创建好的对象（会影响logger的正常工作）
exp = Experiment(**opt)
# 给logger加handler
# 子进程的log->MemoryHandler->OutputHandler-> queue <-ExpOutputDialog.receive->QTextEdit
# log通过queue在进程间传递，主线程通过thread接收queue中的新消息
from logging.handlers import MemoryHandler
handler = MemoryHandler(capacity=1024,
flushLevel=logging.INFO,
target=OutputHandler(q))
exp.logger.addHandler(handler)
exp.run(
visualize_steps=visualize_steps, # should each learning step be shown?
visualize_learning=visualize_learning, # show policy / value function?
visualize_performance=visualize_performance) # show performance runs?
exp.plot()
def __init__(self, domain, representation, policy, steps=100000):
opt = {}
opt["domain"] = domain
# Agent
opt["agent"] = Q_Learning(representation=representation,
policy=policy,
discount_factor=domain.discount_factor,
initial_learn_rate=0.1,
learn_rate_decay_mode="boyan",
boyan_N0=100,
lambda_=0.)
opt["checks_per_policy"] = 10
opt["max_steps"] = steps
opt["num_policy_checks"] = 20
experiment = Experiment(**opt)
experiment.run()
self.policy = opt["agent"].policy
self.domain = domain
