dqn.py 文件源码

python
阅读 34 收藏 0 点赞 0 评论 0

项目:atari-rl 作者: brendanator 项目源码 文件源码
def __init__(self, name, inputs, conv_outputs, reward_scaling, config):
    with tf.variable_scope(name):
      hidden = tf.layers.dense(conv_outputs, 256, tf.nn.relu, name='hidden')

      value = tf.layers.dense(hidden, 1)
      self.value = tf.squeeze(
          inputs.alive * reward_scaling.unnormalize_output(value),
          axis=1,
          name='value')

      actions = tf.layers.dense(hidden, config.num_actions, name='actions')
      self.policy = tf.nn.softmax(actions, name='policy')
      self.log_policy = tf.nn.log_softmax(actions, name='log_policy')

      # Sample action from policy
      self.greedy_action = tf.squeeze(
          tf.multinomial(self.log_policy, num_samples=1),
          axis=1,
          name='greedy_action')
评论列表
文章目录


问题


面经


文章

微信
公众号

扫码关注公众号