Lessons
1Use a tiny example to debug the learning loop35 min read
Calculate one logistic-regression update and identify a sign error.
- →Check a learning update with a hand calculation
- →Use a simple baseline to isolate pipeline defects
2Why precision falls when the world changes35 min read
Calculate precision from prevalence and conditional error rates.
- →Explain precision changes under prevalence shift
3Offline improvement can miss the product objective35 min read
Separate an offline prediction gain from a causal user-outcome claim.
- →Diagnose disagreement between offline and online results
- →Separate predictive association from causal impact
4Present a testable ML plan in ten minutes35 min read
Write a concise decision record with a baseline, test, and stop rule.
- →Present a bounded ML experiment with a decision rule
- →Use a simple baseline to isolate pipeline defects
- →Separate predictive association from causal impact
Skills in this course
- 01Check a learning update with a hand calculationCheck a learning update with a hand calculation.
- 02Explain precision changes under prevalence shiftExplain precision changes under prevalence shift.
- 03Diagnose disagreement between offline and online resultsDiagnose disagreement between offline and online results.
- 04Present a bounded ML experiment with a decision rulePresent a bounded ML experiment with a decision rule.
- 05Use a simple baseline to isolate pipeline defectsUse a simple baseline to isolate pipeline defects.
- 06Separate predictive association from causal impactSeparate predictive association from causal impact.