In order to make progress on https://github.com/paul-tqh-nguyen/reuters_topic_labelling/issues/22 without throwing away our EEAP model implementation (it'd be nice to keep it around for posterity to demonstrate that it functions, though doesn't perform well on our dataset), it'd be useful to abstract out what is generally useful (which is a lot of it, e.g. all the interfaces for training, validating, saving out models, etc.).
paul-tqh-nguyen
commented
4 years ago
In order to make progress on https://github.com/paul-tqh-nguyen/reuters_topic_labelling/issues/22 without throwing away our EEAP model implementation (it'd be nice to keep it around for posterity to demonstrate that it functions, though doesn't perform well on our dataset), it'd be useful to abstract out what is generally useful (which is a lot of it, e.g. all the interfaces for training, validating, saving out models, etc.).
We can do this via an abstract class.
https://docs.python.org/3/library/abc.html provides a way for us to define an abstract class.
This ticket is to abstract out what is generally useful.