• Why interactive graphics for presentation?
  • Enhances knowledge transfer (see New York Times)
• Why interactive graphics for exploration?
  • Identify structure that otherwise goes missing (Tukey 1972)
  • Better understand/diagnose models (Wickham, Cook, & Hofmann 2015).
  • Generate insights faster (iteration time is crucial!)

• Portable (i.e., cross-platform)
• Simple to share
• Encourages composability (i.e., reports, dashboards, etc)
• Enables integration of multiple systems (1 + 1 > 2)

• Great for conveying information (viz is known), but impractical for exploration (viz is unknown).

• The reality: analysts have to learn/juggle many technologies.
• My goal: An R interface that makes 80% of techniques seamless, quick, and easy (w/o knowledge of web technology).
• My contribution: Designed and implemented a linked brushing framework.

     County |   State | TotalVotes| Clinton| Johnson| Stein| Trump| Population|    Area
    --------| --------| ----------|--------|--------|------|------|-----------|--------
1   autauga | alabama |      24661|   0.240|   0.022| 0.004| 0.734|      54571|  594.44
2   baldwin | alabama |      94090|   0.196|   0.026| 0.005| 0.774|     182265| 1589.78
3   barbour | alabama |      10390|   0.467|   0.009| 0.002| 0.523|      27457|  884.88
4      bibb | alabama |       8748|   0.214|   0.014| 0.002| 0.770|      22915|  622.58
5    blount | alabama |      25384|   0.085|   0.013| 0.004| 0.899|      57322|  644.78
6   bullock | alabama |       4701|   0.751|   0.005| 0.002| 0.242|      10914|  622.81
7    butler | alabama |       8685|   0.428|   0.007| 0.001| 0.563|      20947|  776.83
8   calhoun | alabama |      47376|   0.279|   0.024| 0.006| 0.692|     118572|  605.87
9  chambers | alabama |      13778|   0.418|   0.012| 0.003| 0.566|      34215|  596.53
10 cherokee | alabama |      10503|   0.145|   0.014| 0.002| 0.839|      25989|  553.70
# ... with 3,101 more rows

• Is there a relationship between population density and voting preference?

• All examples could classify as focusing, arranging, and/or linking views.
• Cook, Buja, & Swayne 1996: {focusing, arranging, linking} => {find Gestalt, pose queries, make comparisons}

• General purpose:
  • rCharts (Vaidyanathan et al. 2014), ggvis (Chang and Wickham 2016), rBokeh (Hafen et al. 2016), plotly (Sievert el al. 2016)
• Domain specific:
  • leaflet (Cheng et al. 2016), dygraphs (Allaire et al. 2016), networkD3 (Gandrud et. al. 2016), DiagrammeR (Iannone et al. 2016), d3scatter (Cheng 2016)
• Most approaches neglect/ignore linking views and animation.
  • ggvis - limited linking/animation (requires a shiny (Chang 2016) server).
  • animint (Hocking et al. 2016) limited linking/animation by extending ggplot2's grammar of graphics implementation (Wickham et al. 2017).

• plotly (Sievert el al. 2016) borrows from and improves upon animint.
  • Makes ggplot2 graphics interactive.
  • Hooks into custom ggplot2 geoms.
    • E.g., ggmosaic (Jeppson et al. 2016), geomnet (Tyner et al. 2016), and ggalt (Rudis et al. 2016).
  • Special support for ggmatrix objects (from GGally (Schloerke et al. 2016)).
  • Better support for focusing (zoom/pan) and arranging views.
  • A "non-ggplot2" interface supports plots not supported by ggplot2 (e.g., 3D charts).
  • Supports more complex brushing interactions
    • Make selections via mouse hover, click, click & drag.
    • Dynamically modify brush color (making comparisons).

• Keep adding documentation and examples in the plotly for R book.
• Further advance plotly's support for linking views (without shiny).
• Support for more popular ggplot2 extension packages such as ggrepel and ggraph.
  • Integrating plotly's support for linking tree-structures with ggraph would be particularly interesting.
• Enable interactive arranging of views.