Machine Learning Design Document

Question

I was trying to create a design document for a machine learning project.

From a flow perspective, I know that I would have a training flow, a validation/test flow followed by a production flow.

However, from a design document perspective, I was no able to find any sample template that details out the different sections / subsections that a design document for a Machine Learning project should contain. I believe that the design doc for a ML project should be different from a normal project.

Also, what are the different documents that a ML project should have, from a process perspective?

Any ideas in this direction would be helpful.

Answer 1

By ‘design document’ I assume you are authoring a technical doc in an industry job.

I treat this similarly to an academic journal publication but with less formatting and page restrictions. Ultimately, you want to provide sufficient information such that if Apple hired you tomorrow your replacement could reproduce the work. Further, your colleagues should be able to grasp the problem attacked, solution proposed, and evidence of success. I usually put the yawn-inducing deep stuff in appendix, but suggested sections might include: 0) executive summary 1) background/motivation 2) data sources 3) analysis/algorithm/model 4) validation performance 5) [optional] real world test results/pilot 6) [optional] estimated return on investment 7) [appendix] SQL source code, modeling heuristic, outlier treatment, etc.