Databricks ML Associate cheat sheet
Last-minute reference for the Databricks Machine Learning Associate exam. Remember the balance: the platform section outweighs model development.
Exam facts
| Questions | 48 scored |
| Duration | 90 minutes |
| Fee | 200 USD |
| Validity | 2 years |
| Passing score | Not published by Databricks |
Section weights
| Section | Weight |
|---|---|
| Databricks Machine Learning | 38% |
| Model Development | 31% |
| ML Workflows | 19% |
| Model Deployment | 12% |
Platform concepts
| Concept | What it is for |
|---|---|
| Experiment tracking | Parameters, metrics and artefacts per run, so results are comparable and reproducible |
| Model registry | Versioned stage transitions with an approval trail |
| Feature store | One feature definition shared by training and serving — the fix for skew |
| AutoML | A fast baseline and reviewable code, not a final unexamined answer |
| Environment spec | Library versions recorded with the run; the usual reason a run will not reproduce |
Modelling quick answers
- Random split on a time series → leaks the future; split chronologically
- Feature computed from the target → target leakage; great validation, poor production
- Encoder must match at serving → fit in training and persist
- Rare positive class, 99.7% accuracy → precision, recall, AUC-PR
- Train loss down, validation up → overfitting; regularise or stop early
- Large errors disproportionately costly → RMSE
- Wide variance across CV folds → unstable estimate, not a good model
- Two models within 0.5% → take the simpler one
- Why was this prediction made → per-prediction attribution, not global importance
Deployment
- 60M records overnight, no latency need → batch inference job
- Accuracy declining over a year → data and concept drift
The published-score trap
No passing score is published. Any quoted percentage is a guess.
Night-before checklist
- Four sections; the platform section is the largest at 38%
- Tracking, registry, feature store — what each solves
- Leakage versus skew versus drift
- Check ID and proctoring rules — see exam day
Take the 20-question practice test.