Pytorch-lightning: TypeError: validation_step() takes 3 positional arguments but 4 were given

[LuposX]

🐛 Bug

When running my model I get the error message: TypeError: validation_step() takes 3 positional arguments but 4 were given

Stacktrace:

line 106, in <module> trainer.fit(model)
line 707, in fit self.run_pretrain_routine(model)
line 812, in run_pretrain_routine self.evaluate(model, self.get_val_dataloaders(),self.num_sanity_val_steps, self.testing)
line 234, in evaluate
    test)
line 365, in evaluation_forward
    output = model.validation_step(*args)

To Reproduce

Steps to reproduce the behavior:

1. Install PyTorch-lightning with pip install pytorch-lightning
2. Follow the tutorial from William Falcon
3. Run it
4. See error

Code sample

Here is my validation step:

    def validation_step(self, val_batch, batch_idx):
        x, y = val_batch
        logits = self.forward(x)
        loss = self.cross_entropy_loss(logits, y)
        return {'val_loss': loss}

Expected behavior

No error

Environment

Python version: 3.7
Is CUDA available: Yes
CUDA runtime version: 10.2.89
GPU models and configuration: GPU 0: GeForce GTX 1050 Ti
Nvidia driver version: 442.19
cuDNN version: Could not collect

Versions of relevant libraries:
[pip3] numpy==1.16.1
[conda] blas 1.0 mkl
[conda] mkl 2020.0 166
[conda] mkl-service 2.3.0 py37hb782905_0
[conda] mkl_fft 1.0.15 py37h14836fe_0
[conda] mkl_random 1.1.0 py37h675688f_0
[conda] pytorch 1.4.0 py3.7_cuda101_cudnn7_0 pytorch
[conda] pytorch-ignite 0.4.0.dev20200229 pypi_0 pypi
[conda] pytorch-lightning 0.6.0 pypi_0 pypi
[conda] torchvision 0.4.1 pypi_0 pypi
[conda] torchviz 0.0.1 pypi_0 pypi

Additional context

Related Issue I think:
https://github.com/PyTorchLightning/pytorch-lightning/issues/105

[awaelchli]

This should happen when you return multiple dataloaders. In that case, the signature of validation step should be

def validation_step(self, val_batch, batch_idx, dataset_idx):

Did you copy and run the full MNIST example given in the blog post you linked?

[LuposX]

Yeah, I did copy the full code on the blog. I'll try to add dataset_idx to the validation step once I come home.

[annukkaa]

I've encountered the same problem:

TypeError: validation_step() takes 3 positional arguments but 4 were given

Whether running this code or the full version at the end of the colab

Using:

• python 3.6.9
• pytorch 1.4.0
• pytorch-lightning 0.6.0

[awaelchli]

Looks like these notebooks install master, not 0.6.0.

[LuposX]

If I add dataset_idx to my parameters for the validation_step I get the error:

line 56, in validation_step x, y = val_batch
ValueError: too many values to unpack (expected 2)

[rmrao]

I encountered this - you need to add the @pl.data_loader decorator above all your dataloader functions. The examples sometimes use master and seem to be inconsistent.

[danielgreff]

Thanks for your fix. I had actually seen this in the Github Doc, just did not think to implement it. For me it fixes the error I was having.

[LuposX]

For me, it fixed it too thanks.
Correct me if I am wrong In the PyTorch-lightning version 0.6.0 you need
@pl.data_loader decorator and in the masterversion not?

[rmrao]

I'm not fully clear on this - it looks like the basic mnist example does use the decorator. It's not clear if this example is out of date or if other examples are out of date.

[LuposX]

Should I close this issue? Or is this a bug?

[awaelchli]

It should stay open. The examples must be updated so they can work with the upcoming release 0.7.

[awaelchli]

@williamFalcon bringing this to your attention

[williamFalcon]

@LuposX fixed on 0.7.1.
Try again!

Happy to reopen the issue if it's still there.
Also, the docs are much more clear now!