bug in pytorch_lightning.metrics.functional.auroc

[BeHappyForMe]

the code:

def validation_epoch_end(self, outputs):
        .........
        print(total_y_hat.device)
        print(total_y_true.device)
        print(total_y_hat)
        print(total_y_true)
        print(total_y_hat.shape)
        print(total_y_true.shape)
        auc_score = auroc(total_y_hat, total_y_true)

the output is:

Get data done!
Validation sanity check:  50%|█████     | 1/2 [00:00<00:00,  1.06it/s]

cuda:0
cuda:0
tensor([0.5084, 0.5084, 0.5084,  ..., 0.5084, 0.5084, 0.5084], device='cuda:0')
tensor([0., 0., 0.,  ..., 0., 0., 0.], device='cuda:0')
torch.Size([16384])
torch.Size([16384])
Traceback (most recent call last):
  File "lighting_sales.py", line 443, in <module>
    main(hparams)
  File "lighting_sales.py", line 392, in main
    trainer.fit(model)
  File "/opt/conda/lib/python3.7/site-packages/pytorch_lightning/trainer/trainer.py", line 918, in fit
    self.single_gpu_train(model)
  File "/opt/conda/lib/python3.7/site-packages/pytorch_lightning/trainer/distrib_parts.py", line 176, in single_gpu_train
    self.run_pretrain_routine(model)
  File "/opt/conda/lib/python3.7/site-packages/pytorch_lightning/trainer/trainer.py", line 1076, in run_pretrain_routine
    False)
  File "/opt/conda/lib/python3.7/site-packages/pytorch_lightning/trainer/evaluation_loop.py", line 330, in _evaluate
    eval_results = model.validation_epoch_end(outputs)
  File "lighting_sales.py", line 252, in validation_epoch_end
    auc_score = auroc(total_y_hat, total_y_true)
  File "/opt/conda/lib/python3.7/site-packages/pytorch_lightning/metrics/functional/classification.py", line 817, in auroc
    return _auroc(pred=pred, target=target, sample_weight=sample_weight, pos_label=pos_label)
  File "/opt/conda/lib/python3.7/site-packages/pytorch_lightning/metrics/functional/classification.py", line 766, in new_func
    x, y = func_to_decorate(*args, **kwargs)[:2]
  File "/opt/conda/lib/python3.7/site-packages/pytorch_lightning/metrics/functional/classification.py", line 815, in _auroc
    return roc(pred, target, sample_weight, pos_label)
  File "/opt/conda/lib/python3.7/site-packages/pytorch_lightning/metrics/functional/classification.py", line 553, in roc
    pos_label=pos_label)
  File "/opt/conda/lib/python3.7/site-packages/pytorch_lightning/metrics/functional/classification.py", line 504, in _binary_clf_curve
    torch.tensor([target.size(0) - 1])])
RuntimeError: All input tensors must be on the same device. Received cuda:0 and cpu

[github-actions]

Hi! thanks for your contribution!, great first issue!

[ydcjeff]

That bug is fixed on master. See #2657

[BeHappyForMe]

thank u very much

[awaelchli]

Does it work with master branch? If not, do you use the functional or module interface for the metric?

[edenlightning]

@BeHappyForMe I'm closing this but please open if still experiencing this with master.