Method run_epochs

Image_Classification/src/train.py:338–390

(self, trial: int, epochs: List[int])

        return s

    def run_epochs(self, trial: int, epochs: List[int]) -> None:
        for epoch in epochs:
            start_time = time.time()
            train_loss, train_acc1, train_acc5 = self.epoch_iteration(epoch)
            mem_used = self.track_gpu_memory_usage() if MEM_TRACKING else None
            test_loss, test_acc1, test_acc5 = self.validate()
            end_time = time.time()
            tot_time_epoch = end_time - start_time
            results = {"train_loss": train_loss, "train_acc1": train_acc1, "train_acc5": train_acc5,
                       "test_loss": test_loss, "test_acc1": test_acc1, "test_acc5": test_acc5, "time": tot_time_epoch,
                       "mem_used": mem_used}
            self.on_time_results(results, epoch)
            if isinstance(self.scheduler, StepLR) or isinstance(self.scheduler, CosineAnnealingLR):
                self.scheduler.step()
            total_time = time.time()
            scheduler_string = f" w/ {self.config['scheduler']}" if \
                self.scheduler is not None else ''
            print(
                f"{self.config['optimizer']}{scheduler_string} " +
                f"on {self.config['dataset']}: " +
                f"T {trial + 1}/{self.config['n_trials']} | " +
                f"E {epoch + 1}/{epochs[-1] + 1} Ended | " +
                "E Time: {:.3f}s | ".format(end_time - start_time) +
                "~Time Left: {:.3f}s | ".format(
                    (total_time - start_time) * (epochs[-1] - epoch)),
                "Train Loss: {:.4f}% | Train Acc. {:.4f}% | ".format(
                    train_loss,
                    train_acc1) +
                "Test Loss: {:.4f}% | Test Acc. {:.4f}%".format(
                    test_loss,
                    test_acc1))

            if self.early_stop(train_loss):
                print("Info: Early stop activated.")
                break
            if not self.dist:
                data = {'epoch': epoch + 1,
                        'trial': trial,
                        'config': self.config,
                        'state_dict_network': self.network.state_dict(),
                        'state_dict_optimizer': self.optimizer.state_dict(),
                        'state_dict_scheduler': self.scheduler.state_dict()
                        if self.scheduler is not None else None,
                        'best_acc1': self.best_acc1,
                        'output_filename': Path(self.output_filename).name}
                if epoch % self.save_freq == 0:
                    filename = f'trial_{trial}_epoch_{epoch}.pth.tar'
                    torch.save(data, str(self.checkpoint_path / filename))
                if torch.greater(test_acc1, self.best_acc1):
                    self.best_acc1 = test_acc1
                    torch.save(
                        data, str(self.checkpoint_path / 'best.pth.tar'))
                torch.save(data, str(self.checkpoint_path / 'last.pth.tar'))

    def epoch_iteration(self, epoch: int):
        losses = AverageMeter()