Loss functions

[vikashg]

Hi @ericspod,
Can you please help me with this config file. I am not able to start the training using the usual command

python -m monai.bundle run trainer --config_file configs/train.json

I get the following error. Seems like I am making some error on how the loss function is defined.

Traceback (most recent call last):
  File "/raid/Vikash/Tools/MONAI-DEPLOY/MONAI/monai/utils/module.py", line 218, in instantiate
    return component(**kwargs)
TypeError: __init__() got an unexpected keyword argument 'loss'

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/raid/Vikash/Tools/MONAI-DEPLOY/MONAI/monai/bundle/config_item.py", line 285, in instantiate
    return instantiate(modname, **args)
  File "/raid/Vikash/Tools/MONAI-DEPLOY/MONAI/monai/utils/module.py", line 223, in instantiate
    raise RuntimeError(f"Failed to instantiate '{path}' with kwargs: {kwargs}") from e
RuntimeError: Failed to instantiate 'monai.engines.trainer.SupervisedTrainer' with kwargs: {'max_epochs': 200, 'device': device(type='cuda', index=0), 'train_data_loader': <monai.data.dataloader.DataLoader object at 0x7fcc45972580>, 'network': DenseNet121(
  (features): Sequential(
    (conv0): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)

Can you please look into it. I have been strugging with this loss function inclusion for 2 days now :(

[yiheng-wang-nv]

Hi @vikashg , I thinks it's a naming mistake, the argument of "loss" in monai.engines.trainer.SupervisedTrainer should be: loss_function rather than loss.
the source code is:
https://github.com/Project-MONAI/MONAI/blob/fb976d7c6a2fc01640b3593775395f0a7f27b1db/monai/engines/trainer.py#L142

[ericspod]

That does appear to be the issue, we can look at improving the error output to help identify such things.

[vikashg]

Ok great, changing the key from loss to loss_function does help. However, I am having some difficulty in passing the arguments to the loss function. I am trying to figure this out. But if you have some clues, please let me know

Loading dataset: 100%|██████████| 40/40 [00:11<00:00,  3.50it/s]
2022-08-17 10:33:13.012324: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudart.so.11.0
the effective log level of engine logger or RootLogger is higher than INFO, may not record log, please call `logging.basicConfig(stream=sys.stdout, level=logging.INFO)` to enable it.
Current run is terminating due to exception: 'NoneType' object has no attribute 'shape'
Exception: 'NoneType' object has no attribute 'shape'
Traceback (most recent call last):
  File "/home/gupta/anaconda3/envs/monai-deploy4/lib/python3.8/site-packages/ignite/engine/engine.py", line 840, in _run_once_on_dataset
    self.state.output = self._process_function(self, self.state.batch)
  File "/raid/Vikash/Tools/MONAI-DEPLOY/MONAI/monai/engines/trainer.py", line 210, in _iteration
    _compute_pred_loss()
  File "/raid/Vikash/Tools/MONAI-DEPLOY/MONAI/monai/engines/trainer.py", line 202, in _compute_pred_loss
    engine.state.output[Keys.LOSS] = engine.loss_function(engine.state.output[Keys.PRED], targets).mean()
  File "/home/gupta/anaconda3/envs/monai-deploy4/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
    return forward_call(*input, **kwargs)
  File "/raid/Vikash/Tools/MONAI-DEPLOY/MONAI/monai/losses/dice.py", line 723, in forward
    if len(input.shape) != len(target.shape):
AttributeError: 'NoneType' object has no attribute 'shape'
Engine run is terminating due to exception: 'NoneType' object has no attribute 'shape'
Exception: 'NoneType' object has no attribute 'shape'
Traceback (most recent call last):
  File "/home/gupta/anaconda3/envs/monai-deploy4/lib/python3.8/site-packages/ignite/engine/engine.py", line 753, in _internal_run
    time_taken = self._run_once_on_dataset()
  File "/home/gupta/anaconda3/envs/monai-deploy4/lib/python3.8/site-packages/ignite/engine/engine.py", line 854, in _run_once_on_dataset
    self._handle_exception(e)
  File "/home/gupta/anaconda3/envs/monai-deploy4/lib/python3.8/site-packages/ignite/engine/engine.py", line 464, in _handle_exception
    self._fire_event(Events.EXCEPTION_RAISED, e)
  File "/home/gupta/anaconda3/envs/monai-deploy4/lib/python3.8/site-packages/ignite/engine/engine.py", line 421, in _fire_event
    func(*first, *(event_args + others), **kwargs)
  File "/raid/Vikash/Tools/MONAI-DEPLOY/MONAI/monai/handlers/stats_handler.py", line 179, in exception_raised
    raise e
  File "/home/gupta/anaconda3/envs/monai-deploy4/lib/python3.8/site-packages/ignite/engine/engine.py", line 840, in _run_once_on_dataset
    self.state.output = self._process_function(self, self.state.batch)
  File "/raid/Vikash/Tools/MONAI-DEPLOY/MONAI/monai/engines/trainer.py", line 210, in _iteration
    _compute_pred_loss()
  File "/raid/Vikash/Tools/MONAI-DEPLOY/MONAI/monai/engines/trainer.py", line 202, in _compute_pred_loss
    engine.state.output[Keys.LOSS] = engine.loss_function(engine.state.output[Keys.PRED], targets).mean()
  File "/home/gupta/anaconda3/envs/monai-deploy4/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
    return forward_call(*input, **kwargs)
  File "/raid/Vikash/Tools/MONAI-DEPLOY/MONAI/monai/losses/dice.py", line 723, in forward
    if len(input.shape) != len(target.shape):
AttributeError: 'NoneType' object has no attribute 'shape'
Traceback (most recent call last):
  File "/raid/Vikash/Tools/MONAI-DEPLOY/model-zoo/models/breast-model/scripts/script.py", line 14, in <module>
    trainer.run()
  File "/raid/Vikash/Tools/MONAI-DEPLOY/MONAI/monai/engines/trainer.py", line 58, in run
    super().run()
  File "/raid/Vikash/Tools/MONAI-DEPLOY/MONAI/monai/engines/workflow.py", line 290, in run
    super().run(data=self.data_loader, max_epochs=self.state.max_epochs)
  File "/home/gupta/anaconda3/envs/monai-deploy4/lib/python3.8/site-packages/ignite/engine/engine.py", line 704, in run
    return self._internal_run()
  File "/home/gupta/anaconda3/envs/monai-deploy4/lib/python3.8/site-packages/ignite/engine/engine.py", line 783, in _internal_run
    self._handle_exception(e)
  File "/home/gupta/anaconda3/envs/monai-deploy4/lib/python3.8/site-packages/ignite/engine/engine.py", line 464, in _handle_exception
    self._fire_event(Events.EXCEPTION_RAISED, e)
  File "/home/gupta/anaconda3/envs/monai-deploy4/lib/python3.8/site-packages/ignite/engine/engine.py", line 421, in _fire_event
    func(*first, *(event_args + others), **kwargs)
  File "/raid/Vikash/Tools/MONAI-DEPLOY/MONAI/monai/handlers/stats_handler.py", line 179, in exception_raised
    raise e
  File "/home/gupta/anaconda3/envs/monai-deploy4/lib/python3.8/site-packages/ignite/engine/engine.py", line 753, in _internal_run
    time_taken = self._run_once_on_dataset()
  File "/home/gupta/anaconda3/envs/monai-deploy4/lib/python3.8/site-packages/ignite/engine/engine.py", line 854, in _run_once_on_dataset
    self._handle_exception(e)
  File "/home/gupta/anaconda3/envs/monai-deploy4/lib/python3.8/site-packages/ignite/engine/engine.py", line 464, in _handle_exception
    self._fire_event(Events.EXCEPTION_RAISED, e)
  File "/home/gupta/anaconda3/envs/monai-deploy4/lib/python3.8/site-packages/ignite/engine/engine.py", line 421, in _fire_event
    func(*first, *(event_args + others), **kwargs)
  File "/raid/Vikash/Tools/MONAI-DEPLOY/MONAI/monai/handlers/stats_handler.py", line 179, in exception_raised
    raise e
  File "/home/gupta/anaconda3/envs/monai-deploy4/lib/python3.8/site-packages/ignite/engine/engine.py", line 840, in _run_once_on_dataset
    self.state.output = self._process_function(self, self.state.batch)
  File "/raid/Vikash/Tools/MONAI-DEPLOY/MONAI/monai/engines/trainer.py", line 210, in _iteration
    _compute_pred_loss()
  File "/raid/Vikash/Tools/MONAI-DEPLOY/MONAI/monai/engines/trainer.py", line 202, in _compute_pred_loss
    engine.state.output[Keys.LOSS] = engine.loss_function(engine.state.output[Keys.PRED], targets).mean()
  File "/home/gupta/anaconda3/envs/monai-deploy4/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
    return forward_call(*input, **kwargs)
  File "/raid/Vikash/Tools/MONAI-DEPLOY/MONAI/monai/losses/dice.py", line 723, in forward
    if len(input.shape) != len(target.shape):
AttributeError: 'NoneType' object has no attribute 'shape'

[ericspod]

Your transform sequence doesn't appear to have a ground truth "target" value. This would be represented as a label for example when training segmentation. The complaint seems to be that the "target" value in the engine is None which normally would hold the ground truth value whatever it is. See here how it's done for segmentation.

[vikashg]

Yes so I was doubtful about how I am using the transforms to feed into the training and here is my main confusion.
When it comes to segmentation both labels and images can be loaded using the LoadImaged(keys=['image', 'label']) and the subsequent transforms are applied on both images and mask. But that wont be the case in case of classification that is why I just included only the image key in the. In fact for classification training we get the image label pair from the dataset section in the train.json. So I am trying to figure out how to pass the label to the loss function. I will look into the examples you sent.

[ericspod]

Your label needs to be the class then for each image, there still needs to be a ground truth of some sort which is what "label" represents in these data pipelines.