---
title: "Ensamble Graphics and Dashboards"
output:
  html_document:
    toc: yes
---

```{r global_options, include = FALSE}
knitr::opts_chunk$set(collapse = TRUE)
```

```{r, include = FALSE}
library(lattice)
library(tidyverse)
library(gridExtra)
set.seed(12345)
```

## Ensemble Graphics

Data exploration usually produces many graphics.

Being able to reconstruct these form a command history or notebook is
useful.

But a final report or presentation usually needs a small set of
carefully chosen graphics. These can be some or all of:

- Several plots of the same type of the same variables;
  e.g. histograms with different bin widths.

- Differnent views of the same variables, e.g. density plots and box
  plots.

- Plots of the same type of the same variables for different
  subgroups; small multiples, or trellis displays.

Many other variations are possible.  Such collections are sometimes
called _ensemble graphics_.

- To aid in comparisons, axis ranges encoding choices (color, line
  type, etc.) should be coordinated.

- Annotations can be used to emphasize important features.

- The graphs should be organized so the most important features can be
  perceived pre-attentively.

Collections of graphics can be assembled and arranged within the
graphics system or using the facilities of the report generation
system.

- Within the R graphics system you can use features of `par` for base
  graphics and tools from `gridExtra` for `grid`-based graphics.

- `rmarkdown` and LaTeX also provide ways of arranging graphics.

- Tools for creating poster presentations can also help with arranging
  collections of graphs.

Unwin's Fig 12.1 provides an ensemble graphic for a data set on the
chemical composition of coffee samples:

```{r}
library(ggplot2)
library(GGally)

data(coffee, package = "pgmm")
coffee <- within(coffee, Type <- ifelse(Variety == 1,
                                        "Arabica", "Robusta"))
names(coffee) <- abbreviate(names(coffee), 8)
a <- ggplot(coffee, aes(x = Type)) + geom_bar(aes(fill = Type)) +
    scale_fill_manual(values = c("grey70", "red")) +
    guides(fill = FALSE) + ylab("")
b <- ggplot(coffee, aes(x = Fat, y = Caffine, colour = Type)) +
    geom_point(size = 3) +
    scale_colour_manual(values = c("grey70", "red"))
c <- ggparcoord(coffee[order(coffee$Type), ], columns = 3 : 14,
                groupColumn = "Type", scale = "uniminmax") +
    xlab("") + ylab("") +
    theme(legend.position = "none") +
    scale_colour_manual(values = c("grey", "red")) +
    theme(axis.ticks.y = element_blank(),
          axis.text.y = element_blank())
grid.arrange(arrangeGrob(a, b, ncol = 2, widths = c(1, 2)),
             c, nrow = 2)
```

Some of the issues addressed:

- consistent coloring across the plots;

- removing some redundant labeling

- having the red lines appear on top in the parallel coordinates plot.


## Information Dashboards

[_Dashboards_](https://en.wikipedia.org/wiki/Dashboard_(business)) are
popular in business and arose from work on
[_decision support systems_](https://en.wikipedia.org/wiki/Decision_support_system).

Data visualizations are typically a large component of dashboards.

- [Some dashboards](https://web.archive.org/web/20190124133243/https://www.infosys.com/SiteCollectionImages/healthcare-information-dashboards-lrg.jpg)
  take the metaphor too far by showing dials that take up lots of
  space and distract.

- 3D pie charts and the like are also quite common.

- But many are quite effective, convey a lot of information, and
  emphasize the key items well.

- An example from a
  [paper](https://www.perceptualedge.com/articles/Whitepapers/Formatting_and_Layout_Matter.pdf)
  by Stephen Few:

![](img/dash2.png)

Dashboards often have dynamic or interactive features:

- the data displayed may be updated on a regular basis (e.g. stock
  trade activity, current weather conditions);

- the user may be able to interactively change aspects of the
  visualizations.

Ben Schneiderman's design guidelines:

- overview first;
- then zoom/filter;
- details on demand.

Jenifer Tidwell's classification of useful interactions

- scroll and pan;
- zoom;
- open and close;
- sort and rearrange;
- search and filter.

The computational support needed for data updating and interaction
will vary.

- Some approaches can be handled by a browser's JavaScript engine.

- Other approaches need to communicate with a data base or a server
  process.


## Examples

Tableau provides tools for easily creating dashboards; an
[example](https://public.tableau.com/profile/silas.bergen#!/vizhome/LymesDiseaseinMinnesota/Dashboard1)
on lyme's disease in Minnesota.

Rstudio provides the
[flexdashboard](https://pkgs.rstudio.com/flexdashboard/) framework
for creating dashboards with `rmarkdown`.

<!--
- A [dashboard](https://beta.rstudioconnect.com/jjallaire/htmlwidgets-showcase-storyboard/htmlwidgets-showcase-storyboard.html)
   illustrating `htmlwidgets`.
-->

- A
  [shiny example](https://jjallaire.shinyapps.io/shiny-ggplot2-brushing/)
  illustrating brushing.

- A [`flexdashboard` version](dashexamples/coffee.html) of the coffee
  data example.

## Some Notes

- Dashboards are popular but limiting.

- More extensive articles, like one on [power plants in the
  US](https://www.washingtonpost.com/graphics/national/power-plants/?utm_term=.506ef6f03d7c)
  are often more effective.

- Interactive visualizations are also very popular and effective, but
often interactive features are not used. Some links to discussions on
this can be
found [here](https://flowingdata.com/2017/04/05/interactive-visualization-is-still-alive/)

<!--
https://www.perceptualedge.com/articles/visual_business_intelligence/dd_for_rapid_monitoring.pdf
-->

<!--
https://youtu.be/vJG698U2Mvo
- 
-->