Skip to content

Blank-Page Reconstructions — 12 Drills + Key

For each drill, close the source, recreate the full map, then add one changed-fact branch.

Prompts

  1. Chapter operating spine
  2. Six V's
  3. Analytics horizon
  4. ETL vs ELT
  5. Lake vs swamp
  6. Services layers
  7. ML selector
  8. Source filter
  9. Model evidence split
  10. Deployment loop
  11. Governance ring
  12. Tool selection

Reconstruction key

1. Operating spine

Business need → source choice → ingest + Metadata → DQ/align → hypothesis/model → train/evaluate → communicate → deploy/monitor → iteration.

2. Six V's

Big Data center; Volume/amount, Velocity/speed, Variety/forms, Viscosity/integration difficulty, Volatility/change/useful life, Veracity/trust.

3. Analytics horizon

Descriptive = what/why happened → Predictive = what likely → Prescriptive = what action should influence outcome.

4. ETL vs ELT

ETL: Extract → Transform/Integrate → Load. ELT: Extract → Load/Ingest → Transform/Integrate for use.

5. Lake vs swamp

Source → managed ingest → Metadata/provenance → DQ/profile → relationships/keys → governed access. Remove context/control → swamp.

6. Services layers

Batch=complete history; Speed=current/low latency; Serving=merged/exposed view. Add completeness-latency-complexity trade-off.

7. ML selector

Labeled target → supervised; no labels + pattern discovery → unsupervised; goal/reward feedback → reinforcement.

8. Source filter

Business relevance → origin/meaning → grain/timing → consistency/reliability → privacy/bias → feasibility/value → ingest.

9. Model evidence split

History → train/fit → validate/select/tune → test/generalize → deploy; cross-validation rotates held-out folds.

10. Deployment loop

Useful + feasible → deploy → monitor value/performance → refine → new question/source → retrain/reevaluate.

11. Governance ring

Sourcing, sharing, Metadata, enrichment, access, security/privacy, DQ, visualization standards around the Data Science lifecycle.

12. Tool selection

Parallel structured analytics → MPP; flexible varied-data landing → distributed file; compute near data → in-database; elastic/leased exploration → cloud.

Source: maps synthesize pp. 471–501.

← Recall Key · Classification →