[Trainer] dictionary access to multiple named loggers

[remisphere]

🚀 Feature

When using multiple loggers, being able to give a dict of loggers to the Trainer() instead of a list.

Motivation

I find the current list access when calling a single logger not explicit.

Pitch

trainer = Trainer(
    logger={
        'tensorboard': TensorBoardLogger(...),
        'csv': CSVLogger(...)},
    ...
)

then in the loop:

self.logger['tensorboard'].experiment.add_image(...)

instead of

self.logger[0].experiment.add_image(...)

There could be an ambiguity when using an integer as key, so maybe enforce string-only keys.

cc @Borda @awaelchli @edward-io @ananthsub @rohitgr7 @kamil-kaczmarek @Raalsky @Blaizzy @justusschock @kaushikb11

[awaelchli]

This seems reasonable to me. IMO this should also open up discussions around referencing optimizers by name.

[stale]

This issue has been automatically marked as stale because it hasn't had any recent activity. This issue will be closed in 7 days if no further activity occurs. Thank you for your contributions, Pytorch Lightning Team!