Architectural Decisions for Interactive Viz
summary
Outlines concrete architecture choices for interactive visualization at scale: for millions-to-billions of points use bin-summarise-smooth (Datashader) or WebGL engines like deck.gl and/or send a pre-rendered PNG image from server; for subsecond interaction target 200–500 ms latency using in-memory cubes or tiling, approximate engines like VerdictDB or GPU databases (MapD) to trade freshness and flexibility for speed; handle high cardinality with responsive encodings or facet/division (Facet Dive) and reduce dimensions with embeddings; streaming fits fixed sliding windows (cubism) rather than full-history rendering.
AI-written summary · about
chapters (16)
- 0:00Start
- 5:56Grammar of Interactive Graphics
- 7:14Rendering for Data Scale
- 18:11Interaction Speed
- 19:03Multi Multi Chart Layouts
- 19:15Cross Filtering
- 21:31Approximate Query
- 24:30Caching
- 24:49Use a Faster Database
- 27:43More Responsive Visualization to Space and Data
- 29:04Cardinality
- 31:04Dimensionality Reduction
- 37:14Python Universe
- 39:23Mobile Interfaces
- 40:01Adaptivity
- 44:29Automatic Visualization
related talks
37mVisualizing Persistent Data Structures
48mData Visualization in Practice
32mFrom hot data sources to interactive clients in a hurry
30mFlux for Data Visualization: A Preliminary Manifesto
28mBringing Data to your Client-Side Apps
20mA framework for building data driven interactive front ends
49mRendering without lumpy bits
19mVisualizing cloud architectures in real time with d3.js