dslp/steps.md at main · dslp/dslp

Abstraction: Git-based iterative data science project lifecycle process steps

Key points:

• The Data Science Lifecycle Process (DSLP) is cyclical, not linear; teams jump between Ask, Data, Explore, Experiment, and Model steps as understanding evolves
• Each step has a corresponding GitHub issue type (Ask, Data, Experiment, Model) to track work, link related issues, and capture TLDRs of findings
• Explore branches are never merged to the collaboration branch to avoid repo bloat; critical findings are instead written as documentation updates
• Failed experiments are labeled and closed without merging; successful experiments open a Model issue to handle productionization including refactoring, tests, monitoring, and A/B/rollback plans
• Data validation tests and schema definitions should be written during the Data step to reduce downstream issues

Connections: Dslp · Data Science Lifecycle · Mlops · Experiment Tracking

Source: https://github.com/dslp/dslp/blob/main/steps.md