Passing wrong strings for scheduler interval doesn't throw an error

[finnBsch]

I noticed that when you pass an interval as scheduler parameter that is not 'step' or 'epoch' (for example 'batch') there's no error, it just trains without ever calling the scheduler. Had to look into the source code because I didn't find much info about it in the documentation

[Borda]

@finnBsch that is a good observation, mind sending a PR with adding assert or check/raise with exception?

[finnBsch]

I am not sure where the argument gets assigned.. I just know that there is no check for it in update_learning_rates, however I am sure that it gets assigned somewhere earlier but I cannot find it.

[rohitgr7]

@finnBsch here is the code you need to update: https://github.com/PyTorchLightning/pytorch-lightning/blob/8cb2451303fac5d57c79115697e6426f307e8c2c/pytorch_lightning/trainer/optimizers.py#L97