ModuleNotFoundError: No module named 'thop.count_hooks'

[eps696]

UPD: in new thop versions thop.count_hooks should be replaced by thop.vision.basic_hooks

Name: thop Version: 0.0.31.post2005241907 installed via pip

[eps696]

it appeared (during 3rd phase of training), that new thop versions don't work with this code at all, issuing the error:

Traceback (most recent call last):
  File "lib/search/train.py", line 224, in <module>
    main()
  File "lib/search/train.py", line 157, in main
    train(train_loader_model, model, teacher_model, optimizer, lr_policy, logger, epoch)
  File "lib/search/train.py", line 197, in train
    loss = model.module._loss(input, target)
  File "F:\_neuro\_a\util\SR\AGD\lib\search\model_infer.py", line 272, in _loss
    loss = self._criterion(logit, target)
  File "F:\_neuro\_a\util\SR\AGD\lib\search\model_infer.py", line 257, in _criterion
    y_c_features = self.vgg(x)
  File "C:\Users\eps\AppData\Local\Programs\Python\Python35\lib\site-packages\torch\nn\modules\module.py", line 547, in __call__
    result = self.forward(*input, **kwargs)
  File "C:\Users\eps\AppData\Local\Programs\Python\Python35\lib\site-packages\torch\nn\parallel\data_parallel.py", line 150, in forward
    return self.module(*inputs[0], **kwargs[0])
  File "F:\_neuro\_a\util\SR\AGD\lib\search\util_gan\vgg_feature.py", line 26, in __call__
    x_vgg = self.features(x)
  File "C:\Users\eps\AppData\Local\Programs\Python\Python35\lib\site-packages\torch\nn\modules\module.py", line 547, in __call__
    result = self.forward(*input, **kwargs)
  File "C:\Users\eps\AppData\Local\Programs\Python\Python35\lib\site-packages\torch\nn\modules\container.py", line 92, in forward
    input = module(input)
  File "C:\Users\eps\AppData\Local\Programs\Python\Python35\lib\site-packages\torch\nn\modules\module.py", line 549, in __call__
    hook_result = hook(self, input, result)
  File "C:\Users\eps\AppData\Local\Programs\Python\Python35\lib\site-packages\thop\vision\basic_hooks.py", line 31, in count_convNd
    m.total_ops += torch.DoubleTensor([int(total_ops)])
  File "C:\Users\eps\AppData\Local\Programs\Python\Python35\lib\site-packages\torch\nn\modules\module.py", line 591, in __getattr__
    type(self).__name__, name))
AttributeError: 'Conv2d' object has no attribute 'total_ops'

downgrading to thop==0.0.31.post1912272122 made it work with thop.count_hooks, and without this error. later versions have introduced some fatal breaking changes (tested on pytorch 1.2 and 1.6). would be nice to mention this in pip requirements for the repo.

[ghost]

@chenwydj

[tilmto]

Hi, you can install another thop version:

pip install thop==0.0.31.post1912272122

[eps696]

@tilmto yes it's mentioned in the post above. thanks for confirming this.