ValueError: not enough values to unpack (expected 8, got 7)

[zlsbksdxl]

Error ①:

Traceback (most recent call last):
  File "main.py", line 298, in <module>
    main(args, tlogger)
  File "main.py", line 228, in main
    train_loader, val_loader, model, optimizer, schedule, scaler, amp_context, start_epoch = set_environment(args, tlogger)
ValueError: not enough values to unpack (expected 8, got 7)

This error occurs because the number of returned values from the set_environment() function does not match the number of expected values in the main program. In the main program, the unpacking statement is as follows:

train_loader, val_loader, model, optimizer, schedule, scaler, amp_context, start_epoch = set_environment(args, tlogger)

It expects set_environment() to return 8 values, but when train_loader is None (line 77), only 7 values are returned:

if train_loader is None:
    return train_loader, val_loader, model, None, None, None, None

To fix this issue, add start_epoch to the return statement. The modified code is:

if train_loader is None:
    return train_loader, val_loader, model, None, None, None, None, start_epoch

---

Error ②:

Traceback (most recent call last):
  File "main.py", line 298, in <module>
    main(args, tlogger)
  File "main.py", line 254, in main
    eval_and_save(args, model, val_loader)
  File "/root/miniconda3/envs/PIM/lib/python3.8/site-packages/torch/autograd/grad_mode.py", line 28, in decorate_context
    return func(*args, **kwargs)
TypeError: eval_and_save() missing 1 required positional argument: 'tlogger'

When train_loader is None, which means evaluation mode is used without training, the relevant code in the main program is:

if train_loader is not None:
    tlogger.print("Start Training {} Epoch".format(epoch+1))
    train(args, epoch, model, scaler, amp_context, optimizer, schedule, train_loader)
    tlogger.print()
else:
    tlogger.print("Training loader is None, skipping training.")
    from eval import eval_and_save
    eval_and_save(args, model, val_loader)
    break

This error occurs because the call eval_and_save(args, model, val_loader) is missing the required parameter tlogger. Looking at the eval_and_save function in eval.py, it indeed lacks the tlogger parameter. Adding it as a parameter resolves the issue:

@torch.no_grad()
def eval_and_save(args, model, val_loader, tlogger):
    tlogger.print("Start Evaluating")
    acc, eval_name, eval_acces = evaluate(args, model, val_loader)
    tlogger.print("....BEST_ACC: {} {}%".format(eval_name, acc))
    ### build records.txt
    msg = "[Evaluation Results]\n"
    msg += "Project: {}, Experiment: {}\n".format(args.project_name, args.exp_name)
    msg += "Samples: {}\n".format(len(val_loader.dataset))
    msg += "\n"
    for name in eval_acces:
        msg += "    {} {}%\n".format(name, eval_acces[name])
    msg += "\n"
    msg += "BEST_ACC: {} {}% ".format(eval_name, acc)

    with open(args.save_dir + "eval_results.txt", "w") as ftxt:
        ftxt.write(msg)

The modified code in the main program becomes:

if train_loader is not None:
    tlogger.print("Start Training {} Epoch".format(epoch+1))
    train(args, epoch, model, scaler, amp_context, optimizer, schedule, train_loader)
    tlogger.print()
else:
    tlogger.print("Training loader is None, skipping training.")
    from eval import eval_and_save
    eval_and_save(args, model, val_loader, tlogger)  # Add tlogger as a parameter
    break