Error with metrics with `None` averaging

[BardiaKh]

🐛 Bug

There is an issue when using torch metrics with an averaging method of None. It seems that when the trainer is checking if the metrics are ready, as they are tensors of multiple elements (one for each class) they cannot be interpreted as booleans, hence the process crashes.

To Reproduce

https://colab.research.google.com/drive/1Hil03eDMHynRDNa0dG9SAVuPqmc0w2vo?usp=sharing

Expected behavior

Just continue to the second epoch.

Environment

* CUDA:
	- GPU:
		- Tesla K80
	- available:         True
	- version:           10.2
* Packages:
	- numpy:             1.19.5
	- pyTorch_debug:     False
	- pyTorch_version:   1.9.0+cu102
	- pytorch-lightning: 1.4.6
	- tqdm:              4.62.0
* System:
	- OS:                Linux
	- architecture:
		- 64bit
		- 
	- processor:         x86_64
	- python:            3.7.11
	- version:           #1 SMP Sat Jun 5 09:50:34 PDT 2021

Additional context

None.

[tchaton]

Hey @ BardiaKh,

Would you mind making your notebook public ?

Best,
T.C

[BardiaKh]

Sorry for the inconvenience. It should be ok now.

[carmocca]

@BardiaKh Thanks for reporting the error, i've opened a PR to fix it.

However, keep in mind that logging torchmetrics.Metrics whose compute() result is a multidimensional tensor is not supported with self.log(logger=True) as the logger requires that the resulting tensors have a single element.

So you'll get a different error later:

pytorch_lightning.utilities.exceptions.MisconfigurationException: The metric `tensor([0.5000, 1.0000, 0.5000])` does not contain a single element, thus it cannot be converted to a scalar.