本文整理汇总了Python中baselines.her.util.nn方法的典型用法代码示例。如果您正苦于以下问题：Python util.nn方法的具体用法？Python util.nn怎么用？Python util.nn使用的例子？那么恭喜您, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在类baselines.her.util的用法示例。
在下文中一共展示了util.nn方法的1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: __init__
# 需要导入模块: from baselines.her import util [as 别名]
# 或者: from baselines.her.util import nn [as 别名]
def __init__(self, inputs_tf, dimo, dimg, dimu, max_u, o_stats, g_stats, hidden, layers,
**kwargs):
"""The actor-critic network and related training code.
Args:
inputs_tf (dict of tensors): all necessary inputs for the network: the
observation (o), the goal (g), and the action (u)
dimo (int): the dimension of the observations
dimg (int): the dimension of the goals
dimu (int): the dimension of the actions
max_u (float): the maximum magnitude of actions; action outputs will be scaled
accordingly
o_stats (baselines.her.Normalizer): normalizer for observations
g_stats (baselines.her.Normalizer): normalizer for goals
hidden (int): number of hidden units that should be used in hidden layers
layers (int): number of hidden layers
"""
self.o_tf = inputs_tf['o']
self.g_tf = inputs_tf['g']
self.u_tf = inputs_tf['u']
# Prepare inputs for actor and critic.
o = self.o_stats.normalize(self.o_tf)
g = self.g_stats.normalize(self.g_tf)
input_pi = tf.concat(axis=1, values=[o, g]) # for actor
# Networks.
with tf.variable_scope('pi'):
self.pi_tf = self.max_u * tf.tanh(nn(
input_pi, [self.hidden] * self.layers + [self.dimu]))
with tf.variable_scope('Q'):
# for policy training
input_Q = tf.concat(axis=1, values=[o, g, self.pi_tf / self.max_u])
self.Q_pi_tf = nn(input_Q, [self.hidden] * self.layers + [1])
# for critic training
input_Q = tf.concat(axis=1, values=[o, g, self.u_tf / self.max_u])
self._input_Q = input_Q # exposed for tests
self.Q_tf = nn(input_Q, [self.hidden] * self.layers + [1], reuse=True)