data_viz.Rmd

---
title: "Data Visualization in R"
author: "Jake Riley  Sept. 23, 2020"
output: 
  powerpoint_presentation:
    reference_doc: assets/reference.pptx
    knit: (function(inputFile, encoding) {
      rmarkdown::render(inputFile, encoding = encoding, output_dir = "output")
    })
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(
  echo = TRUE,
  warning = FALSE,
  message = FALSE,
  out.height = "300px",
  cache = TRUE
)
```

## Today's talk
- Intro
- What is ggplot
- Tips & Tricks
- Best Practices
- Try it out

## An intro

- Jake Riley
- Clinical Data Analyst at Children's Hospital of Philadelphia
- Avid package developer
- dogdad

@yake_84


## Before we get started

- this talk is aimed at intermediate `ggplot2` users
- everything is within the `tidyverse` framework & R for Data Science (R4DS)
- the pipe `%>%` is used in many places and allows us to create a sequence of manipulations

`iris %>% head()` is the same as `head(iris)`

- the `+` used with `ggplot()` is another type of pipe

- you can pipe from a `dplyr` sequence into a `ggplot()` sequence

## What is ggplot?
 - grammar of graphics
 - just like every sentence has a **subject, verb, and noun**, every chart has a **coordinate system, geom, and aesthetics**
 - the hope is that we will invent new types of charts


```{r}
econ <- 
  ggplot(economics, aes(date, unemploy))

econ +
  geom_line()

econ +
  geom_point()

econ +
  geom_area(fill = "lightblue") +
  geom_point()

econ +
  geom_area(fill = "lightblue") +
  geom_point() + 
  coord_polar()

  
```

 
```{r}
library(tidyverse)

p <- 
  ggplot(mpg) + 
  geom_count(aes(cty, hwy), alpha = 0.5, color = "navyblue") +
  theme_bw() +
  theme(legend.position = "none")

```
 
 
## an example
```{r grammar, eval = FALSE}
p

p + coord_polar()
```

```{r}
gridExtra::grid.arrange(
  p,
  p + coord_polar(), 
  ncol = 1
)
```

## Demystifying `aes()`

- `aes()` = aesthetics
- dynamic, data driven **variables** go inside the `aes()`
- constant, static **values** go outside
- the first 2 arguments of `aes()` are `x` and `y` and I will mostly omit naming these

## Note the difference
- `geom_point(aes(color = class, size = n), ...)`
- `geom_point(aes(...), color = "blue", size = 2)`

```{r out.width = c("50%", "50%"), fig.show = "hold", echo = FALSE}
df <-
  mpg %>% 
  count(cyl = factor(cyl), cty, hwy)

gridExtra::grid.arrange(
  ggplot(df, aes(x = cty, y = hwy)) + geom_point(aes(color = cyl, size = n)),
  ggplot(df, aes(x = cty, y = hwy)) + geom_point(color = "blue", size = 2),
  nrow = 1
)
```


# Tips & Tricks

## Descending bar charts

The number one things I get asked is how to make a barchart go in descending order. 

```{r}
ggplot(mpg, aes(y = class)) + 
  geom_bar()
```

## Arrange by volume: `fct_infreq()`
```{r}
mpg %>%
  mutate(class = fct_infreq(class)) %>% 
  ggplot(aes(y = class)) + 
  geom_bar()
```

## Arrange in descending order: `fct_rev()`
```{r, }
mpg %>% 
  mutate(class = fct_infreq(class)) %>% 
  ggplot(aes(y = fct_rev(class))) + 
  geom_bar()
```

## Aggregated data: `fct_reorder()`
```{r}
mpg %>% 
  count(class) %>% 
  mutate(class = fct_reorder(class, n, sum))
```

## Aggregated data: `geom_col()`
```{r}
mpg %>% 
  count(class) %>% 
  mutate(class = fct_reorder(class, n, sum)) %>% 
  ggplot(aes(n, class)) +
  geom_col()
```


## Too many bars
```{r}
ggplot(mpg, aes(y = model)) +
  geom_bar()
```

## Too many bars: `fct_lump()`
```{r}
mpg %>%
  mutate(
    manufacturer = 
      fct_lump(manufacturer, 5) %>% 
      fct_infreq() %>% 
      fct_rev()
  ) %>% 
  ggplot(aes(y = manufacturer)) +
  geom_bar()
```

## Sorting by fill
```{r}
ggplot(mpg, aes(y = manufacturer, fill = (cyl == 8))) +
  geom_bar()
```

## Sorting by fill
```{r}
mpg %>% 
  mutate(
    cyl_8 = (cyl == 8),
    manufacturer = fct_reorder(manufacturer, cyl_8, sum)
  ) %>% 
  ggplot(aes(manufacturer, fill = cyl_8)) +
  geom_bar() +
  coord_flip()
```

## Sorting by fill (percent)
```{r}
mpg %>% 
  mutate(
    cyl_8 = (cyl == 8),
    manufacturer = 
      fct_reorder(manufacturer, cyl_8, mean)
  ) %>% 
  ggplot(aes(y = manufacturer, fill = cyl_8)) +
  geom_bar(position = "fill")
```


## `facet_grid()` vs `facet_wrap()`
```{r}
p <-
  mpg %>% 
  filter(class != "2seater", cyl != 5) %>% 
  ggplot(aes(cty, hwy, color = class)) + 
  geom_count(alpha = 0.5) +
  lims(x = c(0, NA), y = c(0, NA)) + 
  # can also use xlim() or scale_x_continuous
  guides(color = FALSE) +
  theme(aspect.ratio = 1)

p
```

## `facet_grid()`: new syntax

```{r}
# this is the new syntax, replaces `facet_grid(~class)`
p + facet_grid(cols = vars(class))
```

## facets: scales

```{r}
# scales allows the x & y to vary
# also "free_x", "free_y"
p + facet_grid(cols = vars(class), scales = "free") 
```

## facets: scales & space

```{r}
p + facet_grid(cols = vars(class), scales = "free", space = "free")
```

## facets: margins

```{r}
p + 
  facet_grid(
    rows = vars(year),
    cols = vars(class),
    margins = TRUE
  )
```

## facet_wrap: # of columns/rows

```{r}
# also nrow
p + facet_wrap(~class, ncol = 3)
```

## facet_wrap: scales

```{r}
# space does not work with facet_wrap()
p + facet_wrap(~class, ncol = 3, scales = "free")
```

## facets: (a + b)

```{r}
# also works with facet_grid
p + facet_wrap(~class + year, nrow = 2)
```

## `scale_*_identity()`

Sometimes I want to have better control over colors & sizes. 

Here, I am hard coding the colors

```{r}
df <-
  mpg %>%
  mutate(category = 
    case_when(
      cty < 14 ~ "coral",
      cty > 19 ~ "turquoise",
      TRUE ~ "grey40"
    )
  )
```

## `scale_color_identity()`

```{r}
ggplot(df, aes(cty, hwy, color = category)) +
  geom_count() +
  scale_color_identity()
```

## `scale_fill_identity()`

```{r}
ggplot(df, aes(cty, fill = category)) +
  geom_dotplot() +
  scale_fill_identity()
```

## Geom defaults

```{r}
# default behavior
ggplot(mtcars, aes(mpg, wt)) + 
  geom_point()

# with updates
update_geom_defaults(
  "point", 
  list(color = "dodgerblue", size = 3)
)

ggplot(mtcars, aes(mpg, wt)) + 
  geom_point()

# revert
update_geom_defaults("point", list(color = "black", size = 2))

ggplot(mtcars, aes(mpg, wt)) + 
  geom_point()
```


## theme default
```{r}
# will apply to all future charts
theme_set(theme_classic(base_size = 14))

ggplot(mtcars, aes(mpg, wt)) +
  geom_point()

ggplot(mtcars, aes(mpg, wt)) +
  geom_point() +
  theme(panel.grid.major.x = element_line(color = "orange"))

# can revert back using this code
theme_set(theme_gray())

ggplot(mtcars, aes(mpg, wt)) +
  geom_point() 

```


# Best practices

## The golden ratio 1:1.6

- Try to give your charts the proportion of a credit card
- Also look this up

```{r, out.width = c("60%", "40%"), fig.show = "hold"}
p <- ggplot(mpg, aes(cty)) + geom_area(stat = "count")

gridExtra::grid.arrange(
  p + coord_fixed(1/10),
  p + theme(aspect.ratio = 1/1.6), # ratio depends on the units
  nrow = 1,
  widths = c(0.7, 0.3)
)
```


## Dealing with spaghetti charts
```{r echo = FALSE}
set.seed(1234)
df <-
  expand.grid(
    group = LETTERS[1:5],
    x = as.Date(paste0("2019-01-", 1:30)),
    KEEP.OUT.ATTRS = FALSE
  ) %>% 
  mutate(
    rand = rnorm(-74:75, mean = 0, sd = 4),
    y = ((row_number()/2) %/% 3) + rand + 10,
    category = factor(ntile(y, 4))
  )
  #mutate(y = row_number() %% 3 + row_number() %% 4 + row_number()/150)

my_theme <-
  theme(
    panel.background = element_rect(fill = "white"),
    panel.grid = element_blank(),
    axis.title = element_blank(),
    aspect.ratio = 5/30,
    legend.position = "bottom"
  )
```
This is one of the most common questions to answer: **change for multiple categories over time**. This often results in a chart like the one here It is hard to read but there are some ways you can help your audience

```{r}
ggplot(df, aes(x, y, color = group)) + 
  geom_line()
```

## Highlight the focus & use an informative title

```{r}
ggplot(df, aes(x, y, group = group)) + 
  geom_line(data = filter(df, group != "B"), color = "grey", size = 1) +
  geom_line(data = filter(df, group == "B"), color = "black", size = 2) +
  labs(title = "Group B is currently in the lead")
```

## Facet the data

```{r}
ggplot(df, aes(x, y, fill = (group == "B"))) + 
  geom_area() +
  facet_grid(cols = vars(group)) +
  scale_fill_manual(values = c("grey", "navyblue"))
```


## Try a heatmap but beware

```{r}
h <-
  ggplot(df, aes(x, fct_reorder(group, y, last), fill = y)) + 
  geom_tile(color = "white") +
  scale_fill_gradient2(
    low = "red", mid = "yellow", high = "darkgreen", midpoint = 25
  ) +
  my_theme +
  labs(title = "An improvement, but not colorblind friendly")

h
```

## not colorblind safe
`remotes::install_github("clauswilke/colorblindr")`

```{r}
colorblindr::c
```


## Not every point needs a color

```{r}
ggplot(df, aes(x, fct_reorder(group, y), fill = category)) + 
  geom_tile(color = "white", size = 0.1) +
  scale_fill_manual(
    values = c("grey90", "grey85", "palegreen3", "seagreen4"),
    labels = c("poor", "good", "very good", "excellent")
  ) +
  my_theme +
  labs(title = "All teams are now performing their best")
```


# Partner Activity: Extensions & Addins

## Partner activity

- left side: focus on extensions
- right side: focus on add-ins
- bonus: my `simplecolors` package

- with your partner, review the code and resources below
- find a function or feature that you think is interesting or useful
- place screenshots here https://bit.ly/2XiG5C7
- you don't need to run the code, you can use images from the vignettes
- we'll share at the end


## Extensions & Addins
- [ggplot extensions](https://www.ggplot2-exts.org/ggiraph.html)
- [ggplot addins](https://github.com/daattali/addinslist)

## Extensions

- [ggradar - spider/radar plots](https://github.com/ricardo-bion/ggradar)
- [gganimate](https://gganimate.com/)
- [ggrepel](https://cran.r-project.org/web/packages/ggrepel/vignettes/ggrepel.html)
- [ggforce](https://cran.rstudio.com/web/packages/ggforce/vignettes/Visual_Guide.html)
- [cowplot](https://cran.r-project.org/web/packages/cowplot/vignettes/introduction.html)
- [more](https://www.ggplot2-exts.org/ggiraph.html)



## `addinslist`:

`addinslist::addinslistAddin()`

```{r eval = FALSE, echo = TRUE}
# install.packages("addinslist")
# install.packages("esquisse")
# install.packages("ggedit")
# install.packages("ggThemeAssist")
# install.packages("colourpicker")

data(iris)
data(mpg)

p <-
  ggplot(mpg, aes(cty, hwy)) + 
  geom_point()
```

## Addins in action
```{r eval = FALSE, echo = TRUE}
# esquisse
  esquisse:::esquisser()
  esquisse:::esquisser(mpg)

# others    
  ggThemeAssist::ggThemeAssistGadget(p)
  ggedit(p)
  colourpicker::colourPicker()
```



## `simplecolors`

https://rjake.github.io/simplecolors/articles/intro.html

```{r}
#devtools::install_github("rjake/simplecolors")
library(simplecolors)
simplecolors::show_colors(labels = FALSE)
```

## `simplecolors`

Three main functions:

`sc()`   `sc_across()`   `sc_*()`

```{r, eval = FALSE}
p <-
  ggplot(mpg, aes(hwy, fill = drv)) +
  geom_density(alpha = 0.3)

p + scale_fill_manual(values = sc("blue3", "red3", "violet3"))
p + scale_fill_manual(values = sc_pink(light = c(1,3,5)))
p + scale_fill_manual(values = sc_across("RTV", light = 4, sat = "bright"))
```

```{r, echo = FALSE}
p <-
  ggplot(mpg, aes(hwy, fill = drv)) +
  geom_density(alpha = 0.3)

p1 <- p + scale_fill_manual(values = sc("blue3", "red3", "violet3"))
p2 <- p + scale_fill_manual(values = sc_pink(light = c(1,3,5)))
p3 <- p + scale_fill_manual(values = sc_across("RTV", light = 4, sat = "bright"))
gridExtra::grid.arrange(p1,p2,p3, nrow = 1)
```


# Appendix

## R4DS

R for Data Science is a book all about the `tidyverse`. It is less "data science-y" and more about data manipulation and visualization. It is free online [here](https://r4ds.had.co.nz/) as well as available for sale.


## Stackoverflow

- try `datapasta` for a minimal reprex 
- include images rather than links
- incorporate `styler`


## Cheatsheet

https://github.com/rstudio/cheatsheets/raw/master/data-visualization-2.1.pdf

## Take care when cropping data

The usual methods to "zoom in" can yield unexpected results when stat_ geoms are used. For example, `geom_boxplot()` calls `stat_boxplot()` and filters out data **before** doing the stats and your boxplot will keep readjusting the quartiles
```{r, echo = FALSE}
p <- 
  ggplot(diamonds, aes("", price)) + 
  geom_boxplot()

p

find_limits <- function(x) {
  floor(ggplot_build(x)$data[[1]][2:4])
}
```

## Use `coord_cartesian()` to zoom in

Do not use `ylim()` or `scale_*_continuous()`
```{r}
# find_limits() is a custom function
bind_rows(
  find_limits(p),
  find_limits(p + ylim(0, 12000)),
  find_limits(p + scale_y_continuous(limits = c(0, 12000))),
  find_limits(p + coord_cartesian(ylim = c(0, 12000)))
)

```

# Answers

## Q1
```{r}
mpg %>% 
  mutate(
    drv = fct_infreq(drv) %>% fct_rev()
  ) %>% 
  ggplot(aes(drv)) +
  geom_bar() + 
  coord_flip()

```

## Q2

```{r}
mpg %>% 
  mutate(
    cyl_8 = (cyl == 8),
    manufacturer = fct_reorder(manufacturer, cyl_8, mean)
  ) %>% 
  ggplot(aes(manufacturer, fill = cyl_8)) +
  geom_bar(position = "fill") +
  coord_flip()
```