ec849adaf4ceb42ed52ca142c839f627c34b9434,slm_lab/agent/algorithm/reinforce.py,Reinforce,train,#Reinforce#,95

Before Change

            logger.debug2(f"Training...")
            // We only care about the rewards from the batch
            rewards = self.sample()["rewards"]
            logger.debug3(f"Length first epi: {len(rewards[0])}")
            logger.debug3(f"Len log probs: {len(self.saved_log_probs)}")
            self.net.optim.zero_grad()
            policy_loss = self.get_policy_loss(rewards)
            loss = policy_loss.data[0]
            policy_loss.backward()

After Change

            self.saved_log_probs = []
            self.entropy = []
            logger.debug(f"Policy loss: {loss}")
            return loss.item()
        else:
            return np.nan

    def calc_policy_loss(self, batch):

Instances