Closes #7. Sorry this PR is a bit large, but I couldn't find any good way to break this up: most of the changes are dependent on one another.
This PR has 2 main goals:
Decouple code for loading and preprocessing data from code used to train the model (see #7).
Once that's done, write regression tests to make sure that model output doesn't change when I make future modifications to the code (tests run on a small subset of the data, to keep them running fairly quickly).
This should make it slightly easier if we want to try different datasets or different models in the future.
Closes #7. Sorry this PR is a bit large, but I couldn't find any good way to break this up: most of the changes are dependent on one another.
This PR has 2 main goals:
This should make it slightly easier if we want to try different datasets or different models in the future.