Actively Remove or Archive Features That are Not Used
13 / 46 •
Training •
This practice was ranked as medium.
Click to read more. • This practice helps to increase
the software quality.
Click to read more.
Click to read more. • This practice helps to increase
the software quality.
Click to read more.
Intent
Avoid technical debt caused by unused features.
Motivation
Features that are no longer used introduce technical debt and clutter. Removing or cleaning unused features from the data pipeline helps concentrate only on promising features, and improves understandability and maintenance.
Applicability
Features should be archived whenever features are manually engineered (and not automatically extracted, e.g. through deep learning).
Description
When features which are no longer used are not removed, they introduce clutter in the processing pipeline.
This is equivalent to not removing dead code in traditional programming.
Keeping the pipeline clean from unused features allows faster experimentation and result interpretation, by focusing only on the most relevant features. It also improves debugging.
When removing features, it is also important to consider coverage: if some features are only rarely present, they are good candidates for removal.
If you opt to not remove unused features, make sure that their documentation reflects this status.
Adoption
Related
- Check that Input Data is Complete, Balanced and Well Distributed
- Assign an Owner to Each Feature and Document its Rationale
Read more
- Hidden Technical Debt in Machine Learning Systems
- Rules of Machine Learning: Best Practices for ML Engineering
13 / 46 •
Training •
This practice was ranked as medium.
Click to read more. • This practice helps to increase
the software quality.
Click to read more.
Click to read more. • This practice helps to increase
the software quality.
Click to read more.