TDD shouldn't be TDDious
summary
Shows concretely how notebook exploration becomes a tested pipeline: extracts a Pokemon notebook into small functions, uses pandas and CSV fixtures, stitches stages with pytest-style fixtures, and compares outputs to a snapshot of model predictions so changes are detectable. Claims: tests make refactoring safe, reveal mismatches between notebook and production code, and enforce determinism via random_state. Demonstrates a model that yields 96% accuracy and encodes a 95% threshold to guard regressions.
AI-written summary · about
chapters (15)
- 0:00TDD in data engineering
- 1:34From test plans to TDD
- 3:03BBC iPlayer and TDD
- 4:47TDD in data science
- 6:15Data in Pokémon example
- 7:59What TDD really means
- 11:13Notebook breakdown with TDD
- 14:10Refactoring with small tests
- 16:42Pipeline testing step by step
- 19:52Power of test suite
- 22:21Model accuracy testing
- 24:54Model testing as game
- 26:41Bad data handling
- 27:47TDD shouldn’t feel burdensome
- 28:09Summary and Q&A
