Feature: Plotting of metrics

[SkafteNicki]

What does this PR do?

Fixes #652
Fixes #1265

This PR adds basic plotting functionality to torchmetrics. Idea is that every modular implementation should have an plot method that can create a basic plot of the metric. All plot methods should have the signature:

def compute(self) -> TypeT:
    ...

def plot(self, val: Optional[TypeT, Sequence[TypeT]])
    val = val | self.compute()
    ...

meaning that if no value is provided, self.compute is called. Additionally, either a single value can be provided or a list of values to plot a metric as it has progressed. Please try executing the examples/plotting for examples.

This PR is meant as an proof-on-concept and when it (hopefully) gets merge we can follow up with adding the functionality to other metrics.

Before submitting

• Was this discussed/approved via a Github issue? (no need for typos and docs improvements)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure to update the docs?
• Did you write any new necessary tests?

PR review

Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.

Did you have fun?

Make sure you had fun coding 🙃

[SkafteNicki]

Executing examples/plotting.py should give something like this:

Single metric value

Single metric value, one per class

Multiple metric values

Multiple metric values, one per class

[justusschock]

Something more general: Do we want to support smoothing? Especially for multiple timepoints this is often done to get a more general idea of the progress instead of spiky individual values (See tensorboard for example).

[SkafteNicki]

Something more general: Do we want to support smoothing? Especially for multiple timepoints this is often done to get a more general idea of the progress instead of spiky individual values (See tensorboard for example).

I would be fine with supporting this :)
We would need to add scipy as an dependency, but thats the only "problem".