Update metrics to use Enum

[SkafteNicki]

🚀 Feature

Motivation

Update metrics package to use Enum where it makes sense. For example:
https://github.com/PyTorchLightning/pytorch-lightning/blob/f78223041283450c98fb34bb23bea6706247d937/pytorch_lightning/metrics/classification/helpers.py#L79-L87
the case (data type) could be made an enum type

class DataType(LightningEnum):
    BINARY = 'binary'
    MULTILABEL = 'multi-label'
    MULTICLASS = 'multi-class'
    MDMC = 'multi-dim multi-class'

Pitch

Alternatives

Additional context

@Borda

[yuntai]

I could give it a shot.

something like this
_check_shape_and_type_consistency(preds: torch.Tensor, target: torch.Tensor) -> Tuple[DataType, int]
& typify all string values to DataType in the metric package?
can you point to a file where DataType should be defined?

[SkafteNicki]

The DataType is propably unique to classification problems and I would therefore put it in pytorch-lightning/pytorch_lightning/metrics/classification/helpers.py file.

[yuntai]

fine with

class DataType(LightningEnum):
    BINARY = 'binary'
    MULTILABEL = 'multilabel'
    MULTICLASS = 'multiclass'
    MDMC = 'mdmc'

?

Otherwise
DataType.from_str('multi-label') == None
DataType.MULTILABEL == 'multi-label'
DataType.MULTILABEL != 'multilabel'

[SkafteNicki]

Closed by PR #5689