Lesson 8 — Communication, Deployment, and Tool Choice
Visualization is part of analytical validity
Each visualization should answer a question or provide insight for a defined audience. Choose the design for the decision—not because it is sophisticated.
A chart can use correct source calculations and still mislead through truncated axes, omitted comparison groups, poor labels, or biased framing. Chapter 14 therefore links visualization to explanatory context, neutral presentation, and storytelling.
Prototype ≠ production
Deploy only when the model: - meets a real business need; - has acceptable statistical/model evidence; - is ethically/technically sound enough; - can be operated at justified cost and latency; - can be monitored and maintained.
After deployment, monitor more than accuracy: source meaning changes, DQ, operational health, adoption, business value, cost, and new requirements can all force refinement.
Architecture/tool purposes
MPP shared-nothing
Partition data and computation across nodes with dedicated resources for high-performance parallel analytics. It also supports in-database analytics close to stored data.
Distributed file-based / Hadoop-style environment
Flexible lower-cost landing/storage for very large structured, semi-structured, and unstructured data. It may be less responsive for some structured interrogative analytics.
MapReduce
Map → Shuffle → Reduce: distribute work, organize/intermediate results, then aggregate/produce the needed result.
In-database analytics
Move computation toward the data to reduce huge data movement.
Cloud / leased capacity
Useful for scalable exploration/prototyping when the organization wants to test value before permanent investment.
Tool rule: requirement first, platform second.
Source: pp. 489–497.