e698b32984f3fea15ef6827a9279cb346fc9227b,slm_lab/agent/algorithm/sac.py,SoftActorCritic,train,#SoftActorCritic#,118

Before Change

        if self.to_train == 1:
            batch = self.sample()
            clock.set_batch_size(len(batch))
            pdparams, v_preds = self.calc_pdparam_v(batch)
            advs, v_targets = self.calc_advs_v_targets(batch, v_preds)
            policy_loss = self.calc_policy_loss(batch, pdparams, advs)  // from actor

After Change

            // forward passes for losses
            states = batch["states"]
            actions = batch["actions"]
            v_preds = self.calc_v(states, net=self.critic_net)
            q1_preds = self.calc_q(states, actions, self.q1_net)
            q2_preds = self.calc_q(states, actions, self.q2_net)
            pdparams = self.calc_pdparam(states)
            action_pd = policy_util.init_action_pd(self.body.ActionPD, pdparams)

Instances