Rapid Battle Cards 16–30
- Predictive vs Operational analytics — estimate likely outcome vs apply analytics to live operations/streams and trigger response.
- Choose source vs Ingest source — decide relevance/reliability/fitness vs obtain/onboard data.
- Metadata vs Data Quality — context/origin/meaning/lineage vs fitness/trust/completeness/consistency.
- Data quality vs Model quality — reliability of evidence/input vs soundness/generalization of analytical model.
- Granularity vs Timeliness — level of detail vs how current/frequent data is.
- Training vs Validation — fit parameters/model vs select/tune and estimate selection error.
- Validation vs Test — influence selection/tuning vs independent final generalization check.
- Over-fitting vs generic underperformance — training-specific noise learning vs poor result that may have many causes.
- Outlier vs Error — unusual observation that may be valid/important vs incorrect observation proven defective.
- Static vs Interactive visualization — fixed presentation vs user exploration/manipulation.
- MPP shared-nothing vs Distributed file-based — partitioned parallel analytical processing vs flexible lower-cost file landing/storage.
- Map vs Reduce — transform/process partitions vs aggregate/produce result after shuffle.
- Business relevance vs Technical feasibility — worth doing/value fit vs can it be built/run/maintained within constraints.
- Privacy at source vs Recombination risk — sensitivity visible in original data vs sensitivity/identity emerges after combining/filtering.
- Technical usage metric vs Learning/value metric — capacity/hotspot/activity health vs insight/model/business outcome evidence.
Last-minute drill
Explain Cards 18, 19, 22, 24, 26, 28, and 29 without using the words on either side. If you cannot, return to the matching deep card.