Don’t Underestimate the Obvious: Murphy’s Law in Real-life Data Science
summary
Models break because data and process details break: a weather model trained only on summer data fails in autumn, HTML scraping collects “for sale” pages as real text, and hidden train set contamination or mismatched environments produce deceptively high lab scores. Recommend concrete fixes: start with simple baselines, validate with synthetic data, enforce specify package versions and type checks, use pandas/Dask or PySpark to scale, and add Pytest-backed checks plus production monitoring.
AI-written summary · about
chapters (7)
- 0:00Introduction to talk
- 1:31Real-life data science challenges
- 4:47Data pitfalls and failures
- 9:18Human factor in modeling
- 13:36External world and market needs
- 17:11Summary and takeaways
- 17:53Q&A on testing in production
related talks
19mLearning from eight years of data science mistakes
48mCommon Data Science Mistakes
41mCan Data & ML really learn from software engineering?
57mGood Practices for Applied Machine Learning
50mAnalytics – Balancing ‘Advanced’ and ‘Actionable’
15mMachine Learning: 5 Additional Metrics to Avoid High‑Performing but Useless Models
22mReal-world applications of machine learning: Beyond the hype
48mThe Missing Step in Moving from Data to Models