当前位置: 首页>>代码示例>>Python>>正文


Python objective.TRPO属性代码示例

本文整理汇总了Python中objective.TRPO属性的典型用法代码示例。如果您正苦于以下问题:Python objective.TRPO属性的具体用法?Python objective.TRPO怎么用?Python objective.TRPO使用的例子?那么恭喜您, 这里精选的属性代码示例或许可以为您提供帮助。您也可以进一步了解该属性所在objective的用法示例。


在下文中一共展示了objective.TRPO属性的2个代码示例,这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞,您的评价将有助于系统推荐出更棒的Python代码示例。

示例1: get_objective

# 需要导入模块: import objective [as 别名]
# 或者: from objective import TRPO [as 别名]
def get_objective(self):
    tau = self.tau
    if self.tau_decay is not None:
      assert self.tau_start >= self.tau
      tau = tf.maximum(
          tf.train.exponential_decay(
              self.tau_start, self.global_step, 100, self.tau_decay),
          self.tau)

    if self.objective in ['pcl', 'a3c', 'trpo', 'upcl']:
      cls = (objective.PCL if self.objective in ['pcl', 'upcl'] else
             objective.TRPO if self.objective == 'trpo' else
             objective.ActorCritic)
      policy_weight = 1.0

      return cls(self.learning_rate,
                 clip_norm=self.clip_norm,
                 policy_weight=policy_weight,
                 critic_weight=self.critic_weight,
                 tau=tau, gamma=self.gamma, rollout=self.rollout,
                 eps_lambda=self.eps_lambda, clip_adv=self.clip_adv)
    elif self.objective in ['reinforce', 'urex']:
      cls = (full_episode_objective.Reinforce
             if self.objective == 'reinforce' else
             full_episode_objective.UREX)
      return cls(self.learning_rate,
                 clip_norm=self.clip_norm,
                 num_samples=self.num_samples,
                 tau=tau, bonus_weight=1.0)  # TODO: bonus weight?
    else:
      assert False, 'Unknown objective %s' % self.objective 
开发者ID:ringringyi,项目名称:DOTA_models,代码行数:33,代码来源:trainer.py

示例2: get_objective

# 需要导入模块: import objective [as 别名]
# 或者: from objective import TRPO [as 别名]
def get_objective(self):
    tau = self.tau
    if self.tau_decay is not None:
      assert self.tau_start >= self.tau
      tau = tf.maximum(
          tf.train.exponential_decay(
              self.tau_start, self.global_step, 100, self.tau_decay),
          self.tau)

    if self.objective in ['pcl', 'a3c', 'trpo', 'upcl']:
      cls = (objective.PCL if self.objective in ['pcl', 'upcl'] else
             objective.TRPO if self.objective == 'trpo' else
             objective.ActorCritic)
      policy_weight = 1.0

      return cls(self.learning_rate,
                 clip_norm=self.clip_norm,
                 policy_weight=policy_weight,
                 critic_weight=self.critic_weight,
                 tau=tau, gamma=self.gamma, rollout=self.rollout,
                 eps_lambda=self.eps_lambda, clip_adv=self.clip_adv,
                 use_target_values=self.use_target_values)
    elif self.objective in ['reinforce', 'urex']:
      cls = (full_episode_objective.Reinforce
             if self.objective == 'reinforce' else
             full_episode_objective.UREX)
      return cls(self.learning_rate,
                 clip_norm=self.clip_norm,
                 num_samples=self.num_samples,
                 tau=tau, bonus_weight=1.0)  # TODO: bonus weight?
    else:
      assert False, 'Unknown objective %s' % self.objective 
开发者ID:itsamitgoel,项目名称:Gun-Detector,代码行数:34,代码来源:trainer.py


注:本文中的objective.TRPO属性示例由纯净天空整理自Github/MSDocs等开源代码及文档管理平台,相关代码片段筛选自各路编程大神贡献的开源项目,源码版权归原作者所有,传播和使用请参考对应项目的License;未经允许,请勿转载。