tidyverse_forcats.Rmd

# 因子型变量 {#tidyverse-forcats}

本章介绍R语言中的因子类型数据。因子型变量常用于数据处理和可视化中，尤其在希望不以字母顺序排序的时候，因子就格外有用。

## 什么是因子

因子是把数据进行**分类**并标记为不同层级(level，有时候也翻译成因子水平， 我个人觉得翻译为层级，更接近它的特性，因此，我都会用层级来描述)的数据对象，他们可以存储字符串和整数。因子类型有三个属性：

- 存储类别的数据类型
- 离散变量
- 因子的层级是有限的，只能取因子层级中的值或缺失(NA)


## 创建因子
```{r forcats-1}
library(tidyverse)
library(palmerpenguins)
```

```{r forcats-2}
income <- c("low", "high", "medium", "medium", "low", "high",  "high")
factor(income)
```
因子层级会自动按照字符串的字母顺序排序，比如`high low medium`。也可以指定顺序，


```{r forcats-3}
factor(income, levels = c("low", "high", "medium") )
```


不属于因子层级中的值, 比如这里因子层只有`c("low", "high")`，那么income中的"medium"会被当作缺省值NA
```{r forcats-4}
factor(income, levels = c("low", "high") )
```

相比较字符串而言，因子类型更容易处理，因此很多函数会自动的将字符串转换为因子来处理，但事实上，这也会造成，不想当做因子的却又当做了因子的情形，最典型的是在R 4.0之前，`data.frame()`中`stringsAsFactors`选项，默认将字符串类型转换为因子类型，但这个默认也带来一些不方便，因此在R 4.0之后取消了这个默认。在tidyverse集合里，有专门处理因子的宏包`forcats`，因此，本章将围绕`forcats`宏包讲解如何处理因子类型变量，更多内容可以参考[这里](https://r4ds.had.co.nz/factors.html)。


```{r forcats-5}
library(forcats)
```


## 调整因子顺序

前面看到因子层级是按照字母顺序排序
```{r forcats-6}
x <- factor(income)
x
```

也可以指定顺序
```{r forcats-7}
x %>% fct_relevel( c("high", "medium", "low"))
```
或者让"medium" 移动到最前面
```{r forcats-8}
x %>% fct_relevel( c("medium"))
```
或者让"medium" 移动到最后面

```{r forcats-9}
x %>% fct_relevel("medium", after = Inf)
```


可以按照字符串第一次出现的次序
```{r forcats-10}
x %>% fct_inorder()
```

按照其他变量的中位数的升序排序
```{r forcats-11}
x %>% fct_reorder(c(1:7), .fun = median)  
```



## 应用 

调整因子层级有什么用呢？

这个功能在ggplot可视化中调整分类变量的顺序非常方便。这里为了方便演示，我们假定有数据框
```{r forcats-12}
d <- tibble(
  x = c("a","a", "b", "b", "c", "c"),
  y = c(2, 2, 1, 5,  0, 3)
  
)
d
```


先画个散点图看看吧

```{r forcats-13}
d %>% 
  ggplot(aes(x = x, y = y)) +
  geom_point()
```

我们看到，横坐标上是a-b-c的顺序。


### fct_reorder()

`fct_reorder()`可以让x的顺序按照x中每个分类变量对应y值的中位数升序排序，具体为

- a对应的y值`c(2, 2)` 中位数是`median(c(2, 2)) = 2`
- b对应的y值`c(1, 5)` 中位数是`median(c(1, 5)) = 3`
- c对应的y值`c(0, 3)` 中位数是`median(c(0, 3)) = 1.5`

因此，x的因子层级的顺序调整为c-a-b
```{r forcats-14}
d %>% 
  ggplot(aes(x = fct_reorder(x, y, .fun = median), y = y)) +
  geom_point()
```

当然，我们可以加一个参数`.desc = TRUE`让因子层级变为降序排列b-a-c

```{r forcats-15}
d %>% 
  ggplot(aes(x = fct_reorder(x, y, .fun = median, .desc = TRUE), y = y)) +
  geom_point()
```

但这样会造成x坐标标签一大串，因此建议可以写`mutate()`函数里
```{r forcats-16}
d %>% 
  mutate(x = fct_reorder(x, y, .fun = median, .desc = TRUE)) %>% 
  ggplot(aes(x = x, y = y)) +
  geom_point()
```


我们还可以按照y值中最小值的大小降序排列
```{r forcats-17}
d %>% 
  mutate(x = fct_reorder(x, y, .fun = min, .desc = TRUE)) %>% 
  ggplot(aes(x = x, y = y)) +
  geom_point()
```


### fct_rev()

按照因子层级的逆序排序
```{r forcats-18}
d %>% 
  mutate(x = fct_rev(x)) %>% 
  ggplot(aes(x = x, y = y)) +
  geom_point()
```



### fct_relevel()
```{r forcats-19}
d %>% 
  mutate(
    x = fct_relevel(x, c("c", "a", "b"))
  ) %>% 

  ggplot(aes(x = x, y = y)) +
  geom_point()
```



## 可视化中应用 
可能没说明白，那就看企鹅柱状图吧
```{r forcats-20}
ggplot(penguins, aes(y = species)) +
  geom_bar()
```


```{r forcats-21}
ggplot(penguins, aes(y = fct_rev(species))) +
  geom_bar()
```





```{r forcats-22a, eval=FALSE}
penguins %>% 
  count(species) %>% 
  pull(species)

penguins %>% 
  count(species) %>% 
  mutate(species = fct_relevel(species, "Chinstrap", "Gentoo", "Adelie")) %>% 
  pull(species)
```



```{r forcats-22}
# Move "Chinstrap" in front, rest alphabetic
ggplot(penguins, aes(y = fct_relevel(species, "Chinstrap"))) +
  geom_bar()
```



```{r forcats-23}
# Use order "Chinstrap", "Gentoo", "Adelie"
ggplot(penguins, aes(y = fct_relevel(species, "Chinstrap", "Gentoo", "Adelie"))) +
  geom_bar()
```


```{r forcats-24}
penguins %>%
  mutate(species = fct_relevel(species, "Chinstrap", "Gentoo", "Adelie")) %>%
  ggplot(aes(y = species)) +
  geom_bar()
```


```{r forcats-25}
ggplot(penguins, aes(y = fct_relevel(species, "Adelie", after = Inf))) +
  geom_bar()
```


```{r forcats-26}
# Use the order defined by the number of penguins of different species
# The order is descending, from most frequent to least frequent

penguins %>%
  mutate(species = fct_infreq(species)) %>%
  ggplot(aes(y = species)) +
  geom_bar()
```


```{r forcats-27}
penguins %>%
  mutate(species = fct_rev(fct_infreq(species))) %>%
  ggplot(aes(y = species)) + 
  geom_bar()
```


```{r forcats-28}
# Reorder based on numeric values
penguins %>%
  count(species) %>%
  mutate(species = fct_reorder(species, n)) %>%
  ggplot(aes(n, species)) + 
  geom_col()
```

## 作业

- 画出的2007年美洲人口寿命的柱状图，要求从高到低排序
```{r forcats-29}
library(gapminder)
gapminder %>%
   filter(
     year == 2007,
     continent == "Americas"
)
```


```{r forcats-30, eval=FALSE, echo = FALSE}
gapminder %>%
  filter( year == 2007, continent == "Americas") %>%
  mutate( country = fct_reorder(country, lifeExp)) %>%
  ggplot(aes(lifeExp, country)) +
  geom_point()
```

- 这是四个国家人口寿命的变化图
```{r forcats-31}
gapminder %>%
  filter(country %in% c("Norway", "Portugal", "Spain", "Austria")) %>%
  ggplot(aes(year, lifeExp)) + geom_line() +
  facet_wrap(vars(country), nrow = 1)
```

- 要求给四个分面排序，按每个国家寿命的中位数
```{r forcats-32, eval=FALSE, echo = FALSE}
gapminder %>%
  filter(country %in% c("Norway", "Portugal", "Spain", "Austria")) %>%
  mutate(country = fct_reorder(country, lifeExp)) %>% # default: order by median
  ggplot(aes(year, lifeExp)) + geom_line() +
  facet_wrap(vars(country), nrow = 1)
```



- 要求给四个分面排序，按每个国家寿命差（最大值减去最小值）

```{r forcats-33, eval=FALSE, echo = FALSE}
gapminder %>%
  filter(country %in% c("Norway", "Portugal", "Spain", "Austria")) %>%
  # order by custom function: here, difference between max and min
  mutate(country = fct_reorder(country, lifeExp, function(x) { max(x) - min(x) })) %>%
  ggplot(aes(year, lifeExp)) + geom_line() +
  facet_wrap(vars(country), nrow = 1)
```



```{r forcats-34, echo = F}
# remove the objects
# rm(list=ls())
rm(d, income, x)
```


```{r forcats-35, echo = F, message = F, warning = F, results = "hide"}
pacman::p_unload(pacman::p_loaded(), character.only = TRUE)
```