Learning from eight years of data science mistakes
summary
Eight years of practical lessons: early models leak when features like deposit variable or undated campus visit act as proxies, leading to spurious predictions; a scheduling feature hits a silent time zones bug because data lack timestamps; an experiment is ruined by an AAA test allocation error; business evidence lives in artisanal spreadsheets so a stoplight meta-metrics indicator surfaces data quality; fixes include pseudocode, shared SQL data dictionaries, code reviews, weekly core team syncs and documenting inputs/outputs for reproducibility.
AI-written summary · about
chapters (5)
- 0:00Start
- 1:35DATA SCIENCE MISTAKES OVER TIME
- 2:01Technical / Analysis
- 4:50Communicating with Devs
- 8:35Communicating with Business Stakeholders
related talks
40mMistakes were made: Data science ten years in
48mCommon Data Science Mistakes
54mThe secret sauce of data science management
22mDon’t Underestimate the Obvious: Murphy’s Law in Real-life Data Science
50mAnalytics – Balancing ‘Advanced’ and ‘Actionable’
20mScale up your job satisfaction, not your software
42mHow to be a 10x Data Scientist
40mBuilding Successful Data Science Projects