PyTorch Lightning调用trainer.test报'DummyTqdmFile'无'encoding'属性错误

问题详情 #

使用PyTorch Lightning训练模型并结合Hyperopt调参时，调用trainer.test()触发如下异常：

Traceback (most recent call last):
 File "/home/amaan/code/DL_Simulation/mtl_former.py", line 302, in <module> train()
 File "/home/amaan/code/DL_Simulation/mtl_former.py", line 111, in train best = fmin(
 File "/home/amaan/.conda/envs/dl/lib/python3.9/site-packages/hyperopt/fmin.py", line 540, in fmin return trials.fmin(
 File "/home/amaan/.conda/envs/dl/lib/python3.9/site-packages/hyperopt/base.py", line 671, in fmin return fmin(
 File "/home/amaan/.conda/envs/dl/lib/python3.9/site-packages/hyperopt/fmin.py", line 586, in fmin rval.exhaust()
 File "/home/amaan/.conda/envs/dl/lib/python3.9/site-packages/hyperopt/fmin.py", line 364, in exhaust self.run(self.max_evals - n_done, block_until_done=self.asynchronous)
 File "/home/amaan/.conda/envs/dl/lib/python3.9/site-packages/hyperopt/fmin.py", line 300, in run self.serial_evaluate()
 File "/home/amaan/.conda/envs/dl/lib/python3.9/site-packages/hyperopt/fmin.py", line 178, in serial_evaluate result = self.domain.evaluate(spec, ctrl)
 File "/home/amaan/.conda/envs/dl/lib/python3.9/site-packages/hyperopt/base.py", line 892, in evaluate rval = self.fn(pyll_rval)
 File "/home/amaan/code/DL_Simulation/mtl_former.py", line 112, in <lambda> fn=lambda x: train_single(x, opt_run_id),
 File "/home/amaan/code/DL_Simulation/mtl_former.py", line 181, in train_single val_loss = model_transformer._training_model(
 File "/home/amaan/code/DL_Simulation/model_training/models/model_transformer.py", line 398, in _training_model final_val_loss = BaseTrainingMethods.training_suffix(
 File "/home/amaan/code/DL_Simulation/model_training/models/model_base.py", line 200, in training_suffix trainer.test(
 File "/home/amaan/.conda/envs/dl/lib/python3.9/site-packages/pytorch_lightning/trainer/trainer.py", line 753, in test return call._call_and_handle_interrupt(
 File "/home/amaan/.conda/envs/dl/lib/python3.9/site-packages/pytorch_lightning/trainer/call.py", line 44, in _call_and_handle_interrupt return trainer_fn(*args, **kwargs)
 File "/home/amaan/.conda/envs/dl/lib/python3.9/site-packages/pytorch_lightning/trainer/trainer.py", line 793, in _test_impl results = self._run(model, ckpt_path=ckpt_path)
 File "/home/amaan/.conda/envs/dl/lib/python3.9/site-packages/pytorch_lightning/trainer/trainer.py", line 986, in _run results = self._run_stage()
 File "/home/amaan/.conda/envs/dl/lib/python3.9/site-packages/pytorch_lightning/trainer/trainer.py", line 1023, in _run_stage return self._evaluation_loop.run()
 File "/home/amaan/.conda/envs/dl/lib/python3.9/site-packages/pytorch_lightning/loops/utilities.py", line 182, in _decorator return loop_run(self, *args, **kwargs)
 File "/home/amaan/.conda/envs/dl/lib/python3.9/site-packages/pytorch_lightning/loops/evaluation_loop.py", line 142, in run return self.on_run_end()
 File "/home/amaan/.conda/envs/dl/lib/python3.9/site-packages/pytorch_lightning/loops/evaluation_loop.py", line 274, in on_run_end self._print_results(logged_outputs, self._stage.value)
 File "/home/amaan/.conda/envs/dl/lib/python3.9/site-packages/pytorch_lightning/loops/evaluation_loop.py", line 552, in _print_results if sys.stdout.encoding is not None:
AttributeError: 'DummyTqdmFile' object has no attribute 'encoding'

注释掉trainer.test()后代码可正常运行，但在调试控制台单独执行该调用无报错。核心代码片段：

@staticmethod
def training_suffix(model, args, output_folder, wandb_logger, datamodule, run):
    # 回调与Trainer初始化代码
    early_stopping, checkpoint_callback, lr_monitor = (
        CallbackStorage.get_default_callbacks(
            output_folder=output_folder, filename="model"
        )
    )
    
    args['epochs'] = 2 if args['debug'] else args['epochs']
    trainer = pl.Trainer(
        fast_dev_run=False,
        max_epochs=args["epochs"],
        callbacks=[early_stopping, checkpoint_callback, lr_monitor],
        accelerator="gpu" if torch.cuda.is_available() else "cpu",
        logger=wandb_logger,
        check_val_every_n_epoch=1,
    )
    
    trainer.fit(model, datamodule=datamodule)
    final_val_loss = float(trainer.callback_metrics.get("val_loss"))
    
    print("Final val_loss: ", final_val_loss)

    trainer.test(
        ckpt_path='best',
        dataloaders=datamodule.val_dataloader(),
    )

    # wandb相关代码
    wandb_reference = {
        'entitiy' : run.entity,
        'project' : run.project,
        'id' : run.id
    }
    with open(os.path.join(output_folder, "wandb_information.json"), "w") as file:
        json.dump(wandb_reference, file)

    wandb.finish()
    return final_val_loss

问题原因 #

Hyperopt在运行时会用DummyTqdmFile重定向sys.stdout以适配进度条，但该对象没有PyTorch Lightning打印测试结果时需要的encoding属性，导致触发AttributeError。调试控制台中单独执行时，sys.stdout未被Hyperopt修改，因此无报错。

解决方案 #

方案1：临时补充缺失的encoding属性 #

在调用trainer.test()前，手动为当前sys.stdout添加encoding属性：

import sys

# 在trainer.test()前插入
if not hasattr(sys.stdout, 'encoding'):
    sys.stdout.encoding = 'utf-8'

trainer.test(
    ckpt_path='best',
    dataloaders=datamodule.val_dataloader(),
)

方案2：禁用测试结果打印 #

通过参数关闭PyTorch Lightning的测试结果自动打印，避免触发属性检查：

# 调用trainer.test()时添加verbose=False
trainer.test(
    ckpt_path='best',
    dataloaders=datamodule.val_dataloader(),
    verbose=False
)

或者初始化Trainer时禁用相关打印：

trainer = pl.Trainer(
    # 原有参数保留
    fast_dev_run=False,
    max_epochs=args["epochs"],
    callbacks=[early_stopping, checkpoint_callback, lr_monitor],
    accelerator="gpu" if torch.cuda.is_available() else "cpu",
    logger=wandb_logger,
    check_val_every_n_epoch=1,
    # 新增：禁用测试阶段的结果打印
    enable_progress_bar=False,
    enable_test_loop_logging=False
)

方案3：升级依赖库 #

检查并升级PyTorch Lightning和Hyperopt到最新稳定版本，官方可能已修复该兼容性问题：

pip install --upgrade pytorch-lightning hyperopt

内容的提问来源于stack exchange，提问作者Amaan Ansari