Machine Learning with Swift

Documentation

Surely, it's better when your tool is so simple; it doesn't require a manual. And we all know the deeply rooted tradition of self-documenting code in the Objective-C community. But in the machine learning domain, code without documentation is often useless. Even when there is a documentation, results often cannot be easily reproduced because some exact values of hyperparameters or other seemingly small details are not known.

So, what exactly should be documented in your machine learning-related code? Most importantly:

• Data sources
• Preprocessing steps
• Combinations of features
• Model hyperparameters
• All tricks of the trade
• Error messages
• Loss functions
• Experiments
• Model checkpoints
• Random number seeds
• Quality metrics

Do not forget ...