Mistakes were made: Data science ten years in
summary
Claims a low deployment success rate and lists concrete anti‑patterns: building diagnostics like Yellowbrick, chasing t-SNE vs UMAP visual signal, relying on active learning or synthetic oversampling that fail in production, and training with lasso then serving with random forest or expensive TensorFlow models. Describes rewriting notebook models into a Kafka microservice, implementing a Go ingestion pipeline, and watching everything skew when March 2020 changes traffic. Notes energy/water costs, vendor‑lock and that many prototypes never actually ship.
AI-written summary · about
chapters (15)
- 0:00Welcome
- 1:11Language models and speech evolution
- 4:45Mistakes in data science
- 9:23Categories of mistakes
- 11:48Flawed ML workflow
- 13:56Yellow Brick pitfalls
- 15:46Active learning missteps
- 18:16Government inspection failure
- 23:28OSHA model rejection
- 26:43Trade model errors
- 29:14Entity resolution challenges
- 31:24Customer routing struggles
- 35:19MLOps and drift disaster
- 36:44Anti-patterns in culture
- 39:20Learning from mistakes
related talks
19mLearning from eight years of data science mistakes
42mHow to be a 10x Data Scientist
46mData-Driven Transformation: Organizational Changes and Leadership Insights
54mFrom the Life of a Data Scientist
54mThe secret sauce of data science management
32mWhy every developer should be a data scientist
40mBuilding Successful Data Science Projects
27mFireside Chat