Automated model documentation (H2O AutoDoc)
One of the important roles a data science team performs in an enterprise setting is documenting the history, attributes, and performance of models that are put into production. At a minimum, model documentation should be part of a data science team's best practices. More commonly in an enterprise setting, thorough model documentation or whitepapers are mandated to satisfy internal and external controls as well as regulatory or compliance requirements.
As a rule, model documentation should be comprehensive enough to allow for the recreation of the model being documented. This entails identifying all data sources, including training and test data characteristics, specifying hardware system components, noting software versions, modeling code, software settings and seeds, modeling assumptions adopted, alternative models considered, performance metrics and appropriate diagnostics, and anything else necessary based on business or regulatory...
