question on SAC for discrete action with temperature loss "alpha"

[dbsxdbsx]

According to paper: SOFT ACTOR-CRITIC FOR DISCRETE ACTION SETTINGS, the loss formula (12) for temperature , there is api_t(s_t) at the very outside of the formula, which is action_probs for discrete actions. But according to the code here:

# from  SAC_Discrete.py 
def calculate_actor_loss(self, state_batch):
        """Calculates the loss for the actor. This loss includes the additional entropy term"""
        action, (action_probabilities, log_action_probabilities), _ = self.produce_action_and_action_info(state_batch)
        qf1_pi = self.critic_local(state_batch)
        qf2_pi = self.critic_local_2(state_batch)
        min_qf_pi = torch.min(qf1_pi, qf2_pi)
        inside_term = self.alpha * log_action_probabilities - min_qf_pi
        policy_loss = action_probabilities * inside_term
        policy_loss = policy_loss.mean()
        log_action_probabilities = torch.sum(log_action_probabilities * action_probabilities, dim=1)
        return policy_loss, log_action_probabilities

# from  SAC.py 
 def calculate_entropy_tuning_loss(self, log_pi):
        """Calculates the loss for the entropy temperature parameter. This is only relevant if self.automatic_entropy_tuning
        is True."""
        alpha_loss = -(self.log_alpha * (log_pi + self.target_entropy).detach()).mean()
        return alpha_loss

Here log_action_probabilities = torch.sum(log_action_probabilities * action_probabilities, dim=1) seems that only log_action_probabilities is affected by action_probabilities, but not with self.target_entropy, why?

[dbsxdbsx]

I've solved it here: yining043/SAC-discrete#2 (comment)