viz.Rmd

---
title: Data visualization
author: Aaron A. King
output:
  html_document:
    toc: yes
    toc_depth: 4
    toc_float:
      collapsed: TRUE
      smooth_scroll: TRUE
    highlight: haddock
    number_sections: FALSE
    df_print: paged
    includes:
      after_body:
      - _includes/main_bottom.html
      - _includes/license.html
bibliography: tutorial.bib
csl: jss.csl
---

-----------------------

```{css echo=FALSE,purl=FALSE}
.folder {
	color: #3333ff; 
    font-weight: bold;
}
```
```{r knitr-opts,include=FALSE,purl=FALSE}
prefix <- "viz"
source("_includes/setup.R",local=knitr::knit_global())
```
```{r prelims,include=FALSE,purl=FALSE}
set.seed(594709947L)
library(ggplot2)
theme_set(theme_bw())
```

## How to use this document.

This is an extremely condensed introduction to **R**'s base graphics and---more importantly---the powerful data-visualization package **ggplot2**, developed by Hadley Wickham.
Run the codes shown and study the outputs to learn about these tools.
When questions are posed, do your best to answer them.

For your convenience, the **R** codes for this document are `r xfun::embed_file("viz.R",text="provided in a script")` which you can download, edit, and run.

## Getting started: **R**'s base graphics

### Transgenic mosquito experiment

Let's load the data on transgenic mosquito survival time.

```{r}
dat <- read.csv("https://kingaa.github.io/R_Tutorial/data/mosquitoes.csv")
```

Let's compare the average lifespan of transgenic vs wildtype mosquitoes from this experiment.
The following split the data into two subsets, one for each genetic type.
```{r}
wt <- subset(dat,type=="wildtype",select=lifespan)
tg <- subset(dat,type=="transgenic",select=-type)
```

Let's try and visualize the data.
```{r}
dat$type <- factor(dat$type)
plot(dat)
op <- par(mfrow=c(1,2))
hist(tg$lifespan,breaks=seq(0,55,by=5),ylim=c(0,40))
hist(wt$lifespan,breaks=seq(0,55,by=5),ylim=c(0,40))
par(op)
```

**Question:** What does the second `par` command accomplish?

Another way to visualize a distribution is via the *empirical cumulative distribution plot*.

```{r}
plot(sort(dat$lifespan),seq(1,nrow(dat))/nrow(dat),type='n')
lines(sort(wt$lifespan),seq(1,nrow(wt))/nrow(wt),type='s',col='blue')
lines(sort(tg$lifespan),seq(1,nrow(tg))/nrow(tg),type='s',col='red')
```
**Question:** What does `type="n"` do in the first line above?

### Mammal body and brain sizes

The data on mammal body and brain sizes is included in the **MASS** package:

```{r}
library(MASS)

plot(mammals)
plot(mammals,log='x')
plot(mammals,log='xy')
plot(mammals$body,mammals$brain,log='xy')
plot(brain~body,data=mammals,log='xy')
```

### Oil production

```{r}
read.csv(
  "https://kingaa.github.io/R_Tutorial/data/oil_production.csv",
  comment.char="#"
) -> oil
head(oil)
summary(oil)
plot(oil)
plot(Gbbl~year,data=oil,subset=region=="North.America",type='l')
lines(Gbbl~year,data=oil,subset=region=="Eurasia",type="l",col='red')

library(tidyr)
library(dplyr)

oil |>
  group_by(year) |>
  summarize(Gbbl=sum(Gbbl)) -> total
plot(Gbbl~year,data=total,type='l')
```

## A systematic approach to visualization: the Grammar of Graphics

Parts of a graphic:

1. ***Data***
1. ***Geometrical object***: point, line, box, bar, density plot, contours, ribbons
1. ***Statistical transformations***: bins, mean, median, quantile, ECDF, identity 
1. ***Aesthetic attributes***: x and y position, color, fill, size, shape, line type, transparency
1. ***Scales***: map the data onto the aesthetic attributes
1. A ***coordinate system***: maps x and y position onto the page
1. A ***faceting system***: multiple plots

You construct a graphical visualization by choosing the constituent parts.
This is implemented in the **ggplot2** package.

### References

- [ggplot2.tidyverse.org](https://ggplot2.tidyverse.org/)
- [ggplot2 documentation](https://ggplot2.tidyverse.org/reference/)


## Examples

### Energy production

```{r}
library(readr)
read_csv(
  "https://kingaa.github.io/R_Tutorial/data/energy_production.csv",
  comment="#"
) -> energy

library(ggplot2)

ggplot(data=energy,mapping=aes(x=year,y=TJ,color=region,linetype=source))+
  geom_line()
ggplot(data=energy,mapping=aes(x=year,y=TJ,color=region))+
  geom_line()+
  facet_wrap(~source)
ggplot(data=energy,mapping=aes(x=year,y=TJ,color=source))+
  geom_line()+
  facet_wrap(~region,ncol=2)
```

What can you conclude from the above?
Try plotting these data on the log scale (`scale_y_log10()`).
How does your interpretation change?

```{r}
ggplot(data=energy,mapping=aes(x=year,y=TJ))+
  geom_line()
ggplot(data=energy,mapping=aes(x=year,y=TJ,group=source))+
  geom_line()
```

**Question:** How do you account for the appearance of the two plots immediately above?

```{r}
ggplot(data=energy,mapping=aes(x=year,y=TJ,group=interaction(source,region)))+
  geom_line()
```

**Question:** What does the `group` aesthetic do?

Let's aggregate across regions by year and source of energy.
```{r}
energy |>
  group_by(year,source) |>
  summarize(TJ=sum(TJ)) |>
  ungroup() -> tot

tot |>
  ggplot(aes(x=year,y=TJ,color=source))+
  geom_line()

tot |>
  ggplot(aes(x=year,y=TJ,fill=source))+
  geom_area()
```

Now let's aggregate across years by region and source.

See the [data munging tutorial](./munging.html) for more information on manipulating and reshaping data frames.

```{r}
energy |>
  group_by(region,source) |>
  summarize(TJ=mean(TJ)) |>
  ungroup() -> reg

reg |>
  ggplot(aes(x=region,y=TJ,fill=source))+
  geom_bar(stat="identity")+
  coord_flip()

reg |>
  group_by(region) |>
  mutate(frac = TJ/sum(TJ)) |>
  ungroup() -> reg

reg |>
  ggplot(aes(x=region,y=frac,fill=source))+
  geom_bar(stat="identity")+
  coord_flip()+
  labs(y="fraction of production",x="region")
```

In the above, we first average across years for every region and source.
Then, for each region, we compute the fraction of the total production due to each source.
Finally, we plot the fractions using a barplot.
The `coord_flip` coordinate specification gives us horizontal bars instead of the default vertical bars.
Fancy!

Let's compare fossil fuel production to renewable.
We divide the sources into three types: Carbon-based, Nuclear, and Renewable.
We accomplish this using a "crosswalk" table:
```{r}
data.frame(
  source=c("Coal","Gas","Oil","Nuclear","Hydro","Other Renewables"),
  source1=c("Carbon","Carbon","Carbon","Nuclear","Renewable","Renewable")
) |>
  right_join(energy,by="source") -> energy

energy |>
  group_by(source1,region,year) |>
  summarize(TJ = sum(TJ)) |>
  ungroup() -> x

x |>
  ggplot(aes(x=year,y=TJ,fill=source1))+
  geom_area()+
  facet_wrap(~region,ncol=2)+
  labs(fill="source")

x |>
  ggplot(aes(x=year,y=TJ,fill=source1))+
  geom_area()+
  facet_wrap(~region,scales="free_y",ncol=2)+
  labs(fill="source")

x |>
  group_by(source1,year) |>
  summarize(TJ = sum(TJ)) |>
  ungroup() -> y

y |>
  ggplot(aes(x=year,y=TJ,fill=source1))+
  geom_area()+
  labs(fill="source")
```

--------------------------

### Exercise

Ask a question regarding one of the datasets shown here and devise a visualization to answer it.

--------------------------

Produced with **R** version `r getRversion()`.

--------------------------