After changing to pytorch-lightning 1.5.2, omitting the argument of trainer.test() does not work.

[Keiku]

In case of pytorch-lightning 1.5.2, omission of argument of trainer.test() does not work.

        trainer.fit(model, datamodule)
        trainer.test()

If it is below, it worked fine. Is this the expected behavior?

        trainer.fit(model, datamodule)
        trainer.test(model, datamodule.test_dataloader())

I'm currently refactoring the following:

Keiku/PyTorch-Lightning-CIFAR10: "Not too complicated" training code for CIFAR-10 by PyTorch Lightning https://github.com/Keiku/PyTorch-Lightning-CIFAR10

[rohitgr7]

yes, you need to pass the datamodule. I think this was the case after v1.5 and not just v1.5.2.

[Keiku]

Thank you for your answering. I will continue to catch up on pytorch lighting.

[rohitgr7]

also, you can just pass in

trainer.test(model, datamodule=datamodule)

not need to call .test_dataloader() explicitly.