Week 5 - Adding More Functionality and Refactoring

Modularising functions used in the training file, ensuring reused functions are put into a seperate utility function library. Doing it this way improves code maintability and readability for collaborative development. The functions migrated include the function responsible for creating the data loaders, the one responsible for the training and validation loops, and the one responsible for writing the metrics and loss values to .json and .csv files.
Adding functionality to load a pre-trained model into the training file, bypassing the training and validation loop and starting at the evaluation process. This is good to test how pretrained models perform on previously unseen test sets.

One challenge faced was ensuring that when the training and validation loops were migrated, their functionality. This taught me that when primative types are passed into functions, Python creates local copies so any modifications don’t affect the original variables. However, since objects like the model and optimizer are passed into functions as references to objects in memory, any modifications in the functions affect the original object. This realisation helped me solve a ZeroDivisionError.