plotly fundamentals

class: bottom, left
background-image: url(plotly-for-r.svg)
background-size: contain

# Welcome

## Thanks for coming!

---
class: inverse, center
background-image: url(../your-turn.jpeg)
background-size: contain

## Your turn

.pull-left[
(1) Install the required software <https://workshops.cpsievert.me/20171118/>

(2) Run the docker container so that RStudio appears when you visit <http://localhost:8787> (in Chrome or Firefox, please!)

(3) In the R console, enter 
```r
browseURL("~/day1/index.html")
```

If you see [this popup](popup.png), press "Try Again".

<img src="popup.png" />
]

.pull-right[
(4) Go to Tools -> Global Options and configure your pane layout [like this](panes.png).

(5) Share (w/ your neighbor) 3 things you're hoping to take away from this workshop (share them via Slack if you like!)
]

---
class: middle

<h3 align="center"> About me </h3>

* Maintainer of plotly's R package (for nearly 3 years!)
    * Before that: animint, LDAvis, pitchRx, rdom

* PhD in statistics from Iowa State
    * Thesis: [Interfacing R with Web Technologies for Interactive Statistical Graphics and Computing with Data](https://search.proquest.com/openview/3a91971f82fd4af20a78bebb079f5035/1.pdf?pq-origsite=gscholar&cbl=18750&diss=y)

* Founder & CEO, [Sievert Consulting LLC](https://consulting.cpsievert.me/)
    * Currently looking for more projects!

---
## About you

<https://plot.ly/ggplot2/geom_density/>

---
## Same data, another look

<https://plot.ly/r/parallel-coordinates-plot/>

---
## Is there anything in particular that you would like covered?

* I'm most interested in **linking and animating views without shiny**. Already pretty well acquainted with R/RStudio/Plotly, but hoping to learn how to use Plotly more efficiently and extend my current usage with animation, crosstalk, etc.  I use Plotly most often within flexdashboard applications, so most interested in non-Shiny options.
* I want to learn how to **animate my plotly graphs/charts.**
* More **shiny** dev
* **Creating documents with data visualizations in Markdown files**. I've been working to get better with Markdown, but if there's a way to include interactive visualizations and/or animations in Markdown files that can be sent to readers that would be great. I work in consulting so I often need to send reports to executives and including interactive data visualizations in those reports would be incredibly useful for me.
* Looks good! Stoked for the class. For me, I **don't use rmarkdown** and rarely use ggplot2 - so those could be dropped from the list.
* Looks like all of Day1 currently would be old news. Wouldn't **half a day of R/RStudio/ggplot2** suffice for your audience?

---
## Is there anything else that we can help you with at the workshop?

* I'd like to know how plotly performs with displaying **large data** sets
* Plotting large amounts of **time series** data efficiently
* Plotly custom **controls and mapping** options. 
* Learn how to apply **interactive filters** in Plotly if that's supported (e.g. click on a single bar within a bar chart on the left which filters the heat-map to the right). 
* More advanced visualization techniques such as a chord diagram or a sunburst 
  * Check out [sunburstR](https://github.com/timelyportfolio/sunburstR)
* Suggestions/hints on how to create clean, **good code** would be especially appreciated.
  * Check out [tidyverse](https://www.tidyverse.org/)
* **Extended breakout session** time for help on specific project i am bringing

---
## Is there anything else that we can help you with at the workshop?

.footnote[
### I'll try my best -- 2 days is not enough!
### PLEASE PLEASE PLEASE stop me to clarify and/or ask questions
]

---
## A minimal bar chart

* Every **plotly** chart is powered by [plotly.js](https://github.com/plotly/plotly.js), plus some extra R/JS magic 🎩 🐰.
* How/why did `plot_ly()` draw a bar chart? What if we want something different?

```r
library(plotly)
plot_ly(x = c("A", "B"), y = 1:2)
```

---
class: inverse, center
background-image: url(../your-turn.jpeg)
background-size: contain

## Your turn

In your R console enter:

```r
file.edit('~/day1/your-turn.R')
```

Work through the comments/code/questions in this R script.

Feel free to work with your neighbor and ask me questions!

.footnote[
If you have question(s) about it **rmarkdown**, now would be a good time to ask (we won't have time to cover it together).

**PS.** Did you know [the workshop slides](index.Rmd) were created with **rmarkdown**?
]

---

The **scatter** trace type is quite general. It provides the foundation for many charts (e.g., [polygons](https://plot.ly/ggplot2/geom_polygon/), [ribbons](https://plot.ly/ggplot2/geom_ribbon/), [filled areas](https://plot.ly/r/filled-area-plots/), etc), extends to different coordinate systems (e.g, [scatter3d](https://plot.ly/r/3d-line-plots/), [scattergeo](https://plot.ly/r/bubble-maps/), and [scattermapbox](https://plot.ly/r/scattermapbox/)), and rendering systems (e.g., [scattergl](https://plot.ly/r/webgl-vs-svg/))

```r
subplot(shareY = TRUE,
  plot_ly(x = 1:2, y = 1:2),
  plot_ly(x = 1:2, y = 1:2, mode = "lines"),
  plot_ly(x = 1:2, y = 1:2, mode = "markers+lines"),
  plot_ly(x = 1:2, y = 1:2, text = 1:2, mode = "text"),
  plot_ly(x = 1:2, y = 1:2, text = 1:2, mode = "markers+lines+text")
)
```

---
class: inverse, middle, center

# Working with actual data

```r
file.edit('~/day1/demo.R')
```

---
## Which visualization is better?

```r
subplot(shareX = TRUE, nrows = 2,
  plot_ly(logs, x = ~date, y = ~package, z = ~count, type = "heatmap"),
  plot_ly(logs, x = ~date, y = ~count, color = ~package, mode = "lines")
)
```

---
background-image: url(bostock-heer-groups.png)
background-size: contain

## Famous question: which is larger (and by how much)? <font color="#DEE5FF">A</font> or <font color="#FDDFA4">B</font>?

.footnote[
These questions drive at least two influential papers:

* [Cleveland and McGill (1984)](http://info.slis.indiana.edu/~katy/S637-S11/cleveland84.pdf)
* [Bostock and Heer (2010)]()

This figure is from [Data Visualization for Social Science](http://socviz.co/) (highly recommended!) in reference to Bostock and Heer.
]

---
background-image: url(bostock-heer-findings.png)
background-size: contain

## Position is best, especially along common scale and baseline

.footnote[
Figure from [Heer and Bostock (2010)](http://vis.stanford.edu/files/2010-MTurk-CHI.pdf)
]

---
background-image: url(hierarchy.png)
background-size: contain

## A more general guideline from Cleveland and McGill

.footnote[
* Figure from [Data Points: Visualization That Means Something](https://issuu.com/wiley_publishing/docs/data_points_sample_a15b4e87f1b924) by Nathan Yau
]

---
class: center, inverse, middle

# Interactive techniques can aid in these tasks

---
## Again, which is better?

---
background-image: url(https://i.imgur.com/bdd64us.gif)
background-size: contain
class: inverse

## &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; Graphing 5 time series

## &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;&nbsp; &nbsp; &mdash;&mdash;&mdash;&mdash;&mdash;&mdash;&mdash;&mdash;&mdash;&mdash;

---
background-image: url(https://i.imgur.com/bdd64us.gif)
background-size: contain
class: inverse

## &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; Graphing 5 time series

## &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;&nbsp; &nbsp; &mdash;&mdash;&mdash;&mdash;&mdash;&mdash;&mdash;&mdash;&mdash;&mdash;

<h1 align="right"> 1,000 time series!</h2>

---
## With all my `installed.packages()`, yikes!

```r
plot_ly(logz, x = ~date, y = ~count) %>% group_by(package) %>% add_lines(alpha=0.3)
```

---
## Can improve a bit with interaction

```r
library(crosstalk)
SharedData$new(logz, ~package, "Select package(s)") %>% 
  plot_ly(x = ~date, y = ~count) %>% group_by(package) %>% add_lines(alpha=0.3) %>% 
  highlight(dynamic = TRUE, selectize = TRUE, persistent = TRUE)
```

---
## [heatmaply](https://github.com/talgalili/heatmaply#readme) is awesome for visualizing a numeric matrices!

---
background-image: url(https://i.imgur.com/bdd64us.gif)
background-size: contain
class: inverse

## &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; Graphing 1,000 time series

## &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;&nbsp; &nbsp; &mdash;&mdash;&mdash;&mdash;&mdash;&mdash;&mdash;&mdash;&mdash;&mdash;

---
background-image: url(https://i.imgur.com/bdd64us.gif)
background-size: contain
class: inverse

## &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; Graphing 1,000 time series

## &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;&nbsp; &nbsp; &mdash;&mdash;&mdash;&mdash;&mdash;&mdash;&mdash;&mdash;&mdash;&mdash;

<h1 align="right"> 1,000,000 time series!</h2>

---
class: center, middle

# Overview first, then zoom and filter, then details on demand

Ben Shneiderman

.footnote[
Popular information visualization perspective
]

---
class: center, inverse, middle

# Visualization surprise, but don't scale well. Models scale well, but don't surprise

Hadley Wickham

.footnote[
Popular statistical graphics perspective
]

---
class: inverse
background-image: url(../your-turn.jpeg)
background-size: contain

Have a look at some plotly "extension" packages!
  * [heatmaply](https://github.com/talgalili/heatmaply#readme)
  * [iheatmapr](https://github.com/ropensci/iheatmapr#readme)
  * [visdat](http://visdat.njtierney.com/articles/experimental_features.html#interactivity)
  * [vistime](https://github.com/shosaco/vistime#readme)
  * [ggplotgui](https://github.com/gertstulp/ggplotgui#readme)
  
  
.footnote[
**Exercise**: Most of these packages have a function that returns a plotly object (e.g., `heatmaply::heatmaply()`). Use a plotly function to modify/customize the result (e.g., add a title with `plotly::layout()`)

For all CRAN packages that use plotly, see the "Reverse dependencies" section on <https://cran.r-project.org/package=plotly>
]

---
## What about *long* time series?

* *Tens* of thousands points is responsive with *SVG*

```r
y <- sample(c(-1, 1), 1e4, TRUE)
x <- seq(Sys.time(), Sys.time() + length(y) - 1, by = "1 sec")
plot_ly(x = x, y = cumsum(y)) %>% add_lines() %>% rangeslider()
```

---
## What about *long* time series?

* *Hundreds* of thousands time series points is responsive with *WebGL* (but [no rangeslider (yet)](https://community.plot.ly/t/is-there-a-way-to-keep-the-range-slider-while-using-scattergl/4556/3))

```r
y <- sample(c(-1, 1), 1e5, TRUE)
x <- seq(Sys.time(), Sys.time() + length(y) - 1, by = "1 sec")
plot_ly(x = x, y = cumsum(y)) %>% add_lines() %>% toWebGL()
```

---
## What about performance (beyond time-series)?

### SVG vs Canvas, in general

* The *Scalable* in SVG, means scalable in terms of bounding box size.
  * No matter the context, your browser will struggle to render > 30,000 SVG elements.
  * This is why [canvas based elements](https://en.wikipedia.org/wiki/Canvas_element) exist (the difference is similar to pdf vs png)
  
#### Time series doesn't scale well, even in a canvas context

* Time series has performance limitations that other data types don't (this is pretty universal).

### High performance plotly charts

* scattergl -- basically same as scatter (with [limitations](https://github.com/plotly/plotly.js/issues/130)), but [responsive with >1M points](scattergl.R).
* pointcloud -- more restricted than scattergl, but [responsive w/ several million points](pointcloud.R).
* heatmapgl -- [response with >1M cells](heatmapgl.R) in heatmap.

---
## More time series tips

#### Have *lots* of *long* time series?

* Consider [extracting/visualizing features from each series](https://robjhyndman.com/talks/Google-Oct2015-part3.pdf)
  * Some useful packages: [anomalous](https://github.com/robjhyndman/anomalous) and [tscognostics](https://github.com/earowang/tscognostics)
  * See my work on [visualizing pedestrian traffic](https://github.com/cpsievert/pedestrians) with plotly
  
* Consider a tool like [trelliscope](http://ryanhafen.com/blog/trelliscopejs-plotly) for exploring many "groups"

#### Visualization of models/predictions?

* Start with [forecast](https://cran.r-project.org/package=forecast) and/or [mgcv](https://cran.r-project.org/package=mgcv) for model fitting.
* Use a strategy similar to [here](https://plotly-book.cpsievert.me/a-case-study-of-housing-sales-in-texas.html#fig:forecast) to plot forecasts.

#### Is seasonality important?

* Consider "wrapping" your time-series 
  * Wrap (i.e., group) your series by hand (get inspired by this [paper](https://arxiv.org/pdf/1412.6675.pdf))
  * Checkout out [sugrrants](https://github.com/earowang/sugrrants) (`ggplotly()` converts the ggplot2 plots)

---
## Texas housing prices

```r
library(dplyr)
tx <- txhousing %>%
  select(city, year, month, median) %>%
  filter(city %in% c("Galveston", "Midland", "Odessa", "South Padre Island"))
tx
#> # A tibble: 748 x 4
#>         city  year month median
#>        <chr> <int> <int>  <dbl>
#>  1 Galveston  2000     1  95000
#>  2 Galveston  2000     2 100000
#>  3 Galveston  2000     3  98300
#>  4 Galveston  2000     4 111100
#>  5 Galveston  2000     5  89200
#>  6 Galveston  2000     6 108600
#>  7 Galveston  2000     7  99000
#>  8 Galveston  2000     8  96200
#>  9 Galveston  2000     9 104000
#> 10 Galveston  2000    10 118800
#> # ... with 738 more rows
```

---
## Wrap by year, facet by city

```r
ggplot(tx, aes(month, median, group = year)) + geom_line() +
  facet_wrap(~city, ncol = 2)
```

![](index_files/figure-html/unnamed-chunk-10-1.png)

---
## Compare across cities within year *and* across years within city

```r
TX <- SharedData$new(tx, ~year)
p <- ggplot(TX, aes(month, median, group = year)) + geom_line() +
  facet_wrap(~city, ncol = 2)
(gg <- ggplotly(p, tooltip = "year"))
```

---
## Set selection mode and default selections

```r
highlight(gg, "plotly_hover", defaultValues = "2006")
```

---
## Make comparisons with dynamic brush

```r
highlight(gg, dynamic = TRUE, persistent = TRUE, selectize = TRUE)
```

---
## Customize the appearance of selections

```r
highlight(
  gg, dynamic = TRUE, persistent = TRUE, 
  selected = attrs_selected(mode = "markers+lines", marker = list(symbol = "x"))
)
```

---
## Automate queries via animation

```r
p <- ggplot(tx, aes(month, median)) + 
  geom_line(aes(group = year), alpha = 0.2) +
  geom_line(aes(frame = year), color = "red") +
  facet_wrap(~city, ncol = 2)
ggplotly(p)
```

---
class: inverse, center
background-image: url(../your-turn.jpeg)
background-size: contain

## Your turn

Visit [this post](http://ryanhafen.com/blog/trelliscopejs-plotly), replicate the example (no install needed), and use trelliscope to visualize `txhousing` (or, more preferably, your own data!)

.footnote[
## Now would be a good time to ask me about personal projects (I have a 6pm flight tomorrow)!

#### Unfortunately, we won't cover maps in this workshop, but [day 1 of last years workshop](https://plotcon17.cpsievert.me/workshop/day1/#1) covers the topic in depth.
]