ioslides_deck_dc.Rmd

---
title: "So You're Thinking About Animating a Plot in R"
subtitle: "DC R 2019"
author: "Stephanie Kirmer <BR> [www.stephaniekirmer.com](http://www.stephaniekirmer.com) <BR> @[data_stephanie](http://www.twitter.com/data_stephanie)"
date: "November 8, 2019"
output: 
    ioslides_presentation:
        widescreen: true
---

```{r, include = FALSE}
library(gapminder)
library(ggplot2)
library(gganimate)

```

# Follow Along! | https://github.com/skirmer/animating_dataviz

<div class="notes">
Introductions, get the about-me out of the way and give a quick overview

</div>


## Should you animate that plot?

### **Bad Reasons**

- It's neat  
- I want to show off  
- One of my dimensions is time  

### **Good Reasons**

- I have four or more meaningful data dimensions   
- I want to display many levels of a single dimension  

Also: If you need to see all the data points at the same time, don't use animation

<div class="notes">
Balance between encouragement and caution

Time does not necessarily need to be represented in actual temporality    
Talk about what Meaningful Data Dimensions means    
Talk about levels of dimension as factor  

</div>

## Use Case Ideas

<div class="columns-2">

```{r, echo = FALSE, cache=TRUE}

samplep <- ggplot(gapminder, aes(gdpPercap, lifeExp, size = pop, colour = country)) +
  geom_point(alpha = 0.7, show.legend = FALSE) +
  scale_colour_manual(values = country_colors) +
  scale_size(range = c(2, 12)) +
  scale_x_log10() +
  facet_wrap(~continent, ncol = 2) +
  labs(title = 'Year: {frame_time}', x = 'GDP per capita', y = 'life expectancy') +
  transition_time(year) +
  ease_aes('linear')

animate(samplep, width = 450) 
```

<BR>

Thomas Lin Pedersen's example from https://gganimate.com - (@[thomasp85](https://twitter.com/thomasp85))

- Data in meaningful groups changing over long periods of time  
- What does this plot actually tell the viewer?  

</div>

<div class="notes">
TLP is the creator of gganimate- encourage people to check his stuff out   
Discuss this specific plot- what can you learn from it and what is it not good for?   

- trends? y/n   
- comparisons across continents? y/n   
- comparisons within continents across countries? y/n  

</div>

## More Good Examples {.smaller}
<div class="columns-2">

### Multiple animations
- Mikey Harper: https://mikeyharper.uk/animated-plots-with-r/    

### GIS
- Daniela Vasquez: https://d4tagirl.com/2017/05/how-to-plot-animated-maps-with-gganimate  

### Forecasting
- Hao Ye: https://haoye.us/post/2018-09-03-gganimate/  
- Troy @ Peak Maximum [https://peakmaximum.com/](https://peakmaximum.com/2018/12/22/visualizing-prophet-with-gganimate/)   

<BR>

### Shadowing
- Len Kiefer: https://twitter.com/lenkiefer/status/1102274364995719169 

### Model Performance
- Dariya Sydykova: [https://github.com/dariyasydykova](https://github.com/dariyasydykova/open_projects/tree/master/ROC_animation)    

### Groups Over Time with Reveal
- Jovan Lekovic: [https://constantlyplotting.com/](https://constantlyplotting.com/snapshots-young-people-arent-watching-much-tv/)    

</div>

<div class="notes">
Show these examples in realtime - tell people to google for more to get ideas   
It can be hard to imagine your data in an animation, so looking at examples of how other people do it can be v helpful
</div>

## Think About Design

- What information do you want the audience to take away?   

- Movement and speed lead the viewer to interpretation of the data - this can be good or bad  

- Each static frame should be informative alone as well  

(Be open to the idea that you might not need animation to be most effective)  

<div class="notes">
Thinking about your own data and planning ahead is huge

</div>


# The Tools

## Key Packages to Get
```{r}
library(ggplot2) # make plots
library(gganimate) # animate the plots
library(gifski) # render gifs
```

Optional, maybe interesting packages
```{r}
library(transformr) # additional smooth transformations
library(patchwork) # arrange multiple plots in a single layout
library(ggforce) # extended options for ggplot functionality
```

<div class="notes">
To make the example here work, you need the first three for sure- some different transformation options are made available with the fourth

</div>

## Key Concepts

### **Frame**
- Animations are a series of single images (frames) strung together (think of "frames per second")

### **State**
- The current grouping variable being shown. It may be one frame, or a single state representation can stay visible for multiple frames

### **Transition**
- How the visual rendering moves from state to state

### **Enter/Exit**
- How datapoints with no predecessor or follower are handled

<div class="notes">
These are really concepts in the space of animation generally   
Think about old timey cartooning, or flipbooks- one cel at a time, moving past you quick enough that it looks like movement  
What we are doing here is not dissimilar.  
Frame: a measure of time. State: a property of your data.   
Option: look back at the examples from earlier and talk about what the states are?
</div>

# Building the Viz

## Example Project

Using data on displaced persons living in South Africa, show patterns and trends in refugee movement over twenty-plus years.

1. Where do most refugees come from?   

    - These may be places with high instability or political turmoil at a given time  
2. Over time, are the most common countries of origin about the same?   
3. Is the number of refugees coming from the most common countries of origin stable?

Data Source: UNHCR via data.world: [UNHCR's populations of concern residing in South Africa](https://data.world/unhcr/244d728f-06e6-4c25-8eed-69e8fe3bffe0)

<div class="notes">
Example for today's purposes - no special reason
</div>

## Initial Data

Grab dataset of displaced persons living in South Africa
```{r, echo = FALSE}
dtset <- read.csv("data/unhcr-time-series-residing-zaf-csv-1.csv", skip = 1)
names(dtset) <- gsub("\\.", "_", names(dtset))

dtset <- dplyr::filter(dtset, X_date_year > 1995)

knitr::kable(head(dtset, 4))
```


## Mutate Data

Select the top 10 countries of origin for each year, apply some filters.

```{r, message=FALSE, error=FALSE, warning=FALSE, echo=FALSE}
library(magrittr)
library(dplyr)

plotDT <- dtset %>%
  filter(X_country_origin != "Various/Unknown") %>%
  filter(X_population_type == "Refugees (incl. refugee-like situations)") %>%
  group_by(X_date_year) %>%
  mutate(rank = rank(-X_affected, ties.method = "first") * 1) %>%
  ungroup() %>%
  filter(rank <= 10) %>%
  data.frame()

plotDT$X_country_origin <- ifelse(plotDT$X_country_origin == "Serbia and Kosovo (S/RES/1244 (1999))", "Serbia/Kosovo", as.character(plotDT$X_country_origin))

knitr::kable(head(plotDT, 4))
```

## Grouped Bar Plot - Code

Group by country, bar height is affected persons

```{r}
baseplot1 <- ggplot(
  plotDT,
  aes(X_date_year,
    group = X_country_origin,
    fill = as.factor(X_country_origin),
    color = as.factor(X_country_origin))) +
  theme_bw() +
  theme(legend.position = "bottom") +
  geom_bar(aes(y = X_affected), stat = "identity", position = "dodge")

```

## Grouped Bar Plot - Rendered

Ew. This is not effective.


```{r, echo=FALSE, fig.width = 10}
baseplot1
```

## Can Faceting Solve?

No.

```{r, echo=FALSE, fig.width = 10}
baseplot1+facet_grid(X_country_origin~.)
```

## Can Faceting Solve?

Nooooo.

```{r, echo=FALSE, fig.width = 10}
baseplot1+facet_wrap(X_country_origin~.)
```


## What if we try to make a race plot instead?

![Let's make this!](final_race_plot2.gif)


## Reformat the Plot

From this...

```{r}
baseplot1 <- ggplot(plotDT,
  aes(X_date_year,
    group = X_country_origin,
    fill = as.factor(X_country_origin),
    color = as.factor(X_country_origin))) +
  theme_bw() +
  theme(legend.position = "bottom") +
  geom_bar(aes(y = X_affected), stat = "identity", position = "dodge")

```

## Reformat the Plot

To this!

```{r}
baseplot3 <- ggplot(plotDT, 
### <b>
  aes(x=rank, 
  ### </b>
    group = X_country_origin, 
    fill = as.factor(X_country_origin), 
    color = as.factor(X_country_origin)))+
  theme_bw()+
  ### <b>
  theme(legend.position = "none", # Give the plot enough space to have labels
    axis.ticks.y = element_blank(), 
    axis.text.y  = element_blank(), 
    axis.title.y = element_blank(),
    plot.margin = margin(1,1,1,5, "cm"))+
  geom_text(aes(y = 0, label = paste(X_country_origin, " ")), vjust = 0.2, hjust = 1) +
  coord_flip(clip = "off", expand = FALSE, ylim = c(0, 50000)) + # Flip the plot
  ### </b>
  geom_bar(aes(y = X_affected), stat = "identity", position = "identity")+
  ### <b>
  scale_y_continuous(labels = scales::comma) + 
  scale_x_reverse()
  ### </b>

```
</div>

## Render

Now we have the different "frames" all layered on top of each other.
```{r, echo = FALSE, fig.width = 10}
baseplot3
```

# Develop the Animation

## Basic Version

Literally add one more line of code to your ggplot object.

```{r}

animp <- baseplot3 +
### <b>
  transition_states(X_date_year)
### </b>
```

## It's Alive!

<div class="columns-2">
```{r, fig.width = 8, fig.height=4, echo = FALSE, cache = TRUE}
animate(animp, fps = 10, duration = 20, height = 400) 
```

It's nice, but we can do better  

1. Viewer can't tell what the frames represent  
2. Y axis needs labeling, bars need clear labels  
3. It's moving abruptly between frames   

</div>

## Improvements

Solving Problem 1 and 2: Added a descriptive title/label that indicate the year of the frame, label bars

```{r}

animp <- baseplot3 +
### <b>
  geom_text(aes(y = X_affected, 
    label = as.character(X_affected)), 
    color = "black", vjust = 0.2, hjust = -.5)+
  labs(title = "Refugees Residing in South Africa by Origin, {closest_state}"
    , y="Affected Persons")+
### </b>
  transition_states(X_date_year)

```

## Dynamic Labels

```{r, fig.width = 10, fig.height=6, echo = FALSE, cache = TRUE}

animate(animp, fps = 10, duration = 20) 

```

## How Things Move

Solving Problem 3: how do we want the animation elements to move?

Choose exit and enter styles: grow and shrink?

```{r}

animp <- baseplot3 +
  geom_text(aes(y = X_affected, 
    label = as.character(X_affected)), 
    color = "black", vjust = 0.2, hjust = -.5)+
  labs(title = "Refugees Residing in South Africa by Origin, {closest_state}"
    , y="Affected Persons")+
  transition_states(X_date_year)+
### <b>
  enter_grow() +
  exit_shrink()
### </b>

```


## Exit/Enter

It's interesting, but probably not serving the project objectives

```{r, fig.width = 10, fig.height=6, echo = FALSE, cache = TRUE}

animate(animp, fps = 10, duration = 20) 

```


## How More Things Move

Choose easing:   

* movement between positions (moving on page, not exiting or entering)  
* transition speed can change as it moves (functions may be cubic, quartic, etc)    

Here we see `quartic-in-out`

```{r}

animp <- baseplot3 +
  geom_text(aes(y = X_affected, 
    label = as.character(X_affected)), 
    color = "black", vjust = 0.2, hjust = -.5)+
  labs(title = "Refugees Residing in South Africa by Origin, {closest_state}"
  , y="Affected Persons")+
  transition_states(X_date_year)+
### <b>  
  ease_aes('quartic-in-out')
### </b>

```


## Quartic Easing

```{r, fig.width = 10, fig.height=6, echo = FALSE, cache = TRUE}

animate(animp, fps = 10, duration = 20) 

```

## Bouncy Easing

Easing `back-in-out`: Feels a little cartoony- interesting, but again perhaps not what we need

```{r, fig.width = 10, fig.height=6, echo = FALSE, cache = TRUE}

animp <- baseplot3 +
  geom_text(aes(y = X_affected, 
    label = as.character(X_affected)), 
    color = "black", vjust = 0.2, hjust = -.5)+
  labs(title = "Refugees Residing in South Africa by Origin, {closest_state}"
  , y="Affected Persons")+
  transition_states(X_date_year)+
### <b>
  ease_aes('back-in-out')
### </b>

animate(animp, fps = 10, duration = 20) 

```

## How Fast Things Move (Timing)

In addition to entry, exit, and transition easing:  

- Set pace for the states/transitions  
- Ensure no smoothing between the end and restarting  

```{r}

animp <- baseplot3 +
  geom_text(aes(y = X_affected, 
    label = as.character(X_affected)), 
    color = "black", vjust = 0.2, hjust = -.5)+
  labs(title = "Refugees Residing in South Africa by Origin, {closest_state}"
    , y="Affected Persons")+
### <b>  
  transition_states(X_date_year,transition_length = 5, 
      state_length = c(rep(.25, 21), 20), wrap = FALSE)+
### </b>
  ease_aes('linear')+
  enter_fade() +
  exit_fade()

```


## Set Timings

Slower pace feels smoother, and doesn't insinuate that the last frame and the first flow into each other

```{r, fig.width = 10, fig.height=6, echo = FALSE, cache = TRUE}

animate(animp, fps = 10, duration = 20) 

```

## Transition States Drilldown

<!-- {.smaller} -->


<div class="columns-2">

```{r, eval=FALSE}
  transition_states(
    states = X_date_year,
    transition_length = 5, 
    state_length = c(rep(.25, 21), 20),
    wrap = FALSE)+
  
```


**States**: Assign the state unit - here we use the year.  

**Transition length**  

- period of time we use for change from state to state
- slow it down if you want a smooth looking animation   

**State length**  

- period of time where the state stays static  
- here I am making the very last state stay static longer   

**Wrap**: determine whether to apply transition smoothing between the end and restarting   
<BR>
Experiment with these settings to get the look you want!  


## Render the Final Animation

```{r, fig.width = 10, fig.height=6, eval=TRUE, cache = TRUE} 
animate(animp, fps = 10, width = 800, height = 450) 
```

<div class="notes">

Shorter duration: faster animation
More FPS: smoother animation, but bigger file

The Math:
20 second animation
10 frames per second
200 frame animation total

Relative values:
Static state = .25
Transition = 5

22 years = 22 static states
22 transitions, not including start and return to start (no transitions there)

22 x .25 = 5.5
22 x 5 = 110

200/115.5 = 1.73 frames per state-unit

.43 frames per static state = .043 sec per static state
8.65 frames per transition = .865 sec per transition


</div>

## Save the Animation

```{r, eval=FALSE} 
anim_save(filename = "final_race_plot2.gif")
```

## Lessons to Take Away

### **1. Do you need animation?**

Prioritize the transmission of information **effectively**  
Make your plot serve the audience, don't be fancy if it's not helpful  

### **2. Make your animation awesome**

Think carefully about transitions and speed  

- Figure out the in/out transitions that make sense and are not misleading/distracting  
- Give the eye enough time to understand each state  

Get feedback and test your animation on naive viewers  

<div class="notes">

</div>


## Further Links/Reference

https://gganimate.com/

https://ggplot2.tidyverse.org/

https://ggforce.data-imaginist.com/

https://stackoverflow.com/questions/53162821/animated-sorted-bar-chart-with-bars-overtaking-each-other/53163549 (Hat tip to Jon Spring for this awesome starting point for this kind of thing!)

### Even More Examples

- https://twitter.com/WeAreRLadies/status/1123209959020220419

Similar project in D3: https://observablehq.com/@johnburnmurdoch/bar-chart-race-the-most-populous-cities-in-the-world

# Thank you! | [www.stephaniekirmer.com](http://www.stephaniekirmer.com) | @[data_stephanie](http://www.twitter.com/data_stephanie) |  [www.journera.com](http://www.journera.com)