Custom data file for classification seems to be failing

[zippeurfou]

🐛 Bug

When training a classification model on custom data file, the training fails because it expect num_classes

To Reproduce

Use this collab: https://colab.research.google.com/drive/1uamw6SNaOr_4ch24JNxAj2yfgLUKfJqO?usp=sharing Error:

Traceback (most recent call last):
  File "/usr/local/lib/python3.7/dist-packages/lightning_transformers/cli/train.py", line 84, in hydra_entry
    main(cfg)
  File "/usr/local/lib/python3.7/dist-packages/lightning_transformers/cli/train.py", line 78, in main
    logger=logger,
  File "/usr/local/lib/python3.7/dist-packages/lightning_transformers/cli/train.py", line 61, in run
    trainer.fit(model, datamodule=data_module)
  File "/usr/local/lib/python3.7/dist-packages/pytorch_lightning/trainer/trainer.py", line 458, in fit
    self.call_setup_hook(model)
  File "/usr/local/lib/python3.7/dist-packages/pytorch_lightning/trainer/trainer.py", line 1066, in call_setup_hook
    model.setup(stage_name)
  File "/usr/local/lib/python3.7/dist-packages/lightning_transformers/core/model.py", line 88, in setup
    self.configure_metrics(stage)
  File "/usr/local/lib/python3.7/dist-packages/lightning_transformers/task/nlp/text_classification/model.py", line 61, in configure_metrics
    self.prec = Precision(num_classes=self.num_classes)
  File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 948, in __getattr__
    type(self).__name__, name))
AttributeError: 'TextClassificationTransformer' object has no attribute 'num_classes'

Set the environment variable HYDRA_FULL_ERROR=1 for a complete stack trace.

Expected behavior

It should start training.

Environment

check the notebook

[dsindex]

same here~

$ python train.py task=nlp/text_classification dataset=nlp/text_classification/emotion
...
torch.nn.modules.module.ModuleAttributeError: 'TextClassificationTransformer' object has no attribute 'num_classes'

Set the environment variable HYDRA_FULL_ERROR=1 for a complete stack trace.

[SeanNaren]

Thanks guys! This makes sense, I think the right approach to fix this would be to either infer the number of classes from the data (by collecting all unique labels, which I think HF Datasets supports) or to allow the user to pass this in.

[vvssttkk]

this error only for cli when u fixed this, or how I can do this myself?

[vvssttkk]

but, when run predict accept next error

[RaedShabbir]

am running into the same issue, is there any workarounds? ''' TextClassificationTransformer' object has no attribute 'num_classes' '''

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[jmwoloso]

see https://github.com/PyTorchLightning/lightning-transformers/issues/216 and https://github.com/PyTorchLightning/lightning-transformers/issues/215