A qusetion about the code of 'ActorSAC' class in net.py

I'm confused why we use 'logprob = dist.log_prob(a_avg)' instead of 'logprob = dist.log_prob(action)' in line 247 of elegantrl/agents/net.py. I think the latter is consistent to the original paper. Is using the former better in experiment?

````
def get_action_logprob(self, state):
state = self.state_norm(state)
s_enc = self.net_s(state) # encoded state
a_avg, a_std_log = self.net_a(s_enc).chunk(2, dim=1)
a_std = a_std_log.clamp(-16, 2).exp()
dist = Normal(a_avg, a_std)
action = dist.rsample()
action_tanh = action.tanh()
logprob = dist.log_prob(a_avg)
logprob -= (-action_tanh.pow(2) + 1.000001).log() # fix logprob using the derivative of action.tanh()
return action_tanh, logprob.sum(1)
````