2381a50a70559340a0335288d648b4bb9a675588,slm_lab/agent/algorithm/dqn.py,VanillaDQN,train,#VanillaDQN#,124

Before Change

        """
        total_t = util.s_get(self, "aeb_space.clock").get("total_t")
        if (total_t > self.training_min_timestep and total_t % self.training_frequency == 0):
            logger.debug3(f"Training at total_t: {total_t}")
            total_loss = 0.0
            for _b in range(self.training_epoch):
                batch = self.sample()
                batch_loss = 0.0
                for _i in range(self.training_iters_per_batch):
                    with torch.no_grad():
                        q_targets = self.compute_q_target_values(batch)
                        y = q_targets
                    loss = self.net.training_step(batch["states"], y)
                    batch_loss += loss.item()
                batch_loss /= self.training_iters_per_batch
                total_loss += batch_loss
            total_loss /= self.training_epoch
            logger.debug(f"total_loss {total_loss}")
            return total_loss
        else:
            logger.debug3("NOT training")
            return np.nan

    @lab_api
    def body_act(self, body, state):

After Change

            self.body.entropies = []
            logger.debug(f"Loss: {loss}")
            self.last_loss = loss.item()
        return self.last_loss

    @lab_api
    def update(self):
        """Update the agent after training"""

Instances