TST: Use mocks in high-level functions + test Model classes directly #798
Description
Right now we test functions like vak.train.frame_classification.train_frame_classification_model by actually fitting a model. These tests are long running.
I'm wondering if we should instead mock out the call to Trainer.train (or Trainer.test or Trainer.predict as the case may be) and instead test each Model class directly in its tests module, with fake data.
The downside is that we miss out on testing different ways of calling the higher level functions.
But if we want to move towards being more of a domain-specific torch library with support for only a very fixed way of calling these internal functions then that's probably ok.