783e415c29c4a4124ea426c824864a32b0f8ee71,src/garage/tf/policies/gaussian_mlp_policy.py,GaussianMLPPolicy,dist_info_sym,#GaussianMLPPolicy#Any#Any#Any#,192

Before Change

                std_param_var = L.get_output(self._l_std_param, obs_var)
            if self.min_std_param is not None:
                std_param_var = tf.maximum(std_param_var, self.min_std_param)
            if self.std_parametrization == "exp":
                log_std_var = std_param_var
            elif self.std_parametrization == "softplus":
                log_std_var = tf.math.log(
                    tf.math.log(1. + tf.exp(std_param_var)))
            else:
                raise NotImplementedError
            return dict(mean=mean_var, log_std=log_std_var)

    @overrides
    def get_action(self, observation):

After Change

    def dist_info_sym(self, obs_var, state_info_vars=None, name="default"):
        Symbolic graph of the distribution.
        with tf.compat.v1.variable_scope(self._variable_scope):
            mean_var, log_std_var, _, _ = self.model.build(obs_var, name=name)
        return dict(mean=mean_var, log_std=log_std_var)

    def get_action(self, observation):
        Get action from the policy.

Instances