Replace Distributions with tf.Distributions about garage HOT 2 CLOSED

import tensorflow as tf


def entropy(self, obs_var, name=None):
    with tf.name_scope(name, "entropy_sym", [obs_var]):
        _, _, _, dist = self._build_graph(obs_var)
        return dist.entropy()


# old dist maintained in policy
def kl(self, obs_var, name=None):
    with tf.name_scope(name, "kl", [obs_var]):
        _, _, _, dist = self._build_graph(obs_var)
        _, _, _, old_dist = self._build_graph(obs_var, old=True)
        return dist.kl_divergence(old_dist)


# old policy maintained in algorithm
def kl(self, old, obs_var, name=None):
    with tf.name_scope(name, "kl", [obs_var]):
        _, _, _, dist = self._build_graph(obs_var)
        _, _, _, old_dist = old._build_graph(obs_var)
        return dist.kl_divergence(old_dist)


# old dist maintained in policy
def likelihood_ratio(self, obs_var, action_var, name=None):
    with tf.name_scope(name, "likelihood_ratio", [obs_var, action_var]):
        _, _, _, dist = self._build_graph(obs_var)
        _, _, _, old_dist = self._build_graph(obs_var, old=True)

        return tf.exp(
            dist.log_prob(action_var) - old_dist.log_prob(action_var))


# old policy maintained in algorithm
def likelihood_ratio(self, old, obs_var, action_var, name=None):
    with tf.name_scope(name, "likelihood_ratio", [obs_var, action_var]):
        _, _, _, dist = self._build_graph(obs_var)
        _, _, _, old_dist = old._build_graph(obs_var)

        return tf.exp(
            dist.log_prob(action_var) - old_dist.log_prob(action_var))