In The 2020 State of Engineering Practices for Machine Learning we present the findings of our global survey among ML teams.
What was the survey about?
In the survey, we asked ML practitioners to tell us which engineering practices their teams use when building software that uses ML components. The list of practices included in the survey was created through an extensive review of both scientific articles and practitioner blogs, in order to identify which practices they describe and recommend.
Apart from the practices, we also included questions about the effects that each team was able to observe by adopting the practices.
Who answered the survey?
In the first half of 2020, the survey was taken by about 350 practitioners working in research, companies and governmental organisations. In the report, you can read how they are distributed over the globe, over large or small teams, etc.
What did you find?
Among other things, we discovered that:
- Larger teams tend to adopt more practices
- Tech companies lead in practice adoption
- Specific effects (such as traceability) are related to specific sets of practices
The report also provides a full ranking of all practices from most adopted to least adopted.
For more detail, take a look at the report:
The practices from the survey are described in more detail in our online catalogue of ML engineering best practices.
More details about our survey methodology and statistical analysis are available in the scientific article Adoption and Effects of Software Engineering Best Practices in Machine Learning.
To contribute to future versions of our survey report, please take the survey!