30.4 Common Workflow Mistakes Beginners Make

Avoid these

1. Train on all rows, then “test” on the same rows. Scores look perfect; real life fails. Fix: Chapter 31 split.
2. Skip cleaning. Garbage features → confident wrong guesses.
3. Change the label definition halfway. “Late” was 20 minutes, then silently 30.
4. Leak future columns. Using “actual delivery minutes” as a feature to predict lateness (Chapter 32 leakage).
5. Ship without evaluate. Demo on one happy row is not evaluate.

# Bad vs better habit (pseudo checklist)
bad = ["fit on all rows", "no clean step", "label changed quietly"]
better = ["split first", "clean then features", "write label definition in notebook"]
print("Avoid:", bad)
print("Prefer:", better)