INNER CODE UNIT · Python

target

rasbt/machine-learning-book · ch19/cartpole/main.py:85

                    target = r + self.gamma * pred.max()

                target_all = self.model(torch.tensor(s, dtype=torch.float32))[0]
                target_all[a] = target

            batch_states.append(s.flatten())
            batch_targets.append(target_all)
            self._adjust_epsilon()

        self.optimizer.zero_grad()
        pred = self.model(torch.tensor(batch_states, dtype=torch.float32))

        loss = self.loss_fn(pred, torch.stack(batch_targets))
        loss.backward()
        self.optimizer.step()

        return loss.item()

View source record →