22

ggplot2 图形

几何形状与图形语法

Tidyverse 篇
💡

采菊东篱下，悠然见南山。

根据大家投票，觉得ggplot2是最想掌握的技能，我想这就是R语言中最有质感的部分吧。所以，这里专门拿出一节课讲ggplot2，也算是补上之前相关章节数据可视化没讲的内容。

R 
library(tidyverse)

一个有趣的案例

先看一组数据

R 
df <- read_csv("./demo_data/datasaurus.csv")
df

先用dataset分组后，然后计算每组下x的均值和方差，y的均值和方差，以及x，y两者的相关系数，我们发现每组数据下它们几乎都是相等的

R 
df %>%
  group_by(dataset) %>%
  summarise(
    across(everything(), list(mean = mean, sd = sd), .names = "{fn}_{col}")
  ) %>%
  mutate(
    across(is.numeric, round, 3)
  )

如果上面代码不熟悉，可以用相关章节的代码重新表达，也是一样的

R 
df %>%
  group_by(dataset) %>%
  summarize(
    mean_x = mean(x),
    mean_y = mean(y),
    std_dev_x = sd(x),
    std_dev_y = sd(y),
    corr_x_y = cor(x, y)
  )

那么，我们是否能得出结论，每组的数据长的差不多呢？然而，我们画图发现

事实上，每张图都相差很大。所以，这里想说明的是，眼见为实。换句话说，可视化是数据探索中非常重要的部分。本章的目的就是带领大家学习ggplot2基本的绘图技能。

学习目标

图形语法

图形语法 “grammar of graphics” ("ggplot2" 中的gg 就来源于此) 使用图层(layer)去描述和构建图形，下图是ggplot2图层概念的示意图 !图片

这个过程类似我们画一幅水彩画

图片

图形部件

一张统计图形就是从数据到几何形状(geometric object，缩写geom)所包含的图形属性(aesthetic attribute，缩写aes)的一种映射。

  • data: 数据框data.frame (注意，不支持向量vector和列表list类型）
  • aes: 数据框中的数据变量映射到图形属性。什么叫图形属性？就是图中点的位置、形状，大小，颜色等眼睛能看到的东西。什么叫映射？就是一种对应关系，比如数学中的函数b = f(a)就是a和b之间的一种映射关系, a的值决定或者控制了b的值，在ggplot2语法里，a就是我们输入的数据变量，b就是图形属性， 这些图形属性包括：
  • x（x轴方向的位置）
  • y（y轴方向的位置）
  • color（点或者线等元素的颜色）
  • size（点或者线等元素的大小）
  • shape（点或者线等元素的形状）
  • alpha（点或者线等元素的透明度）
  • geoms: 几何形状，确定我们想画什么样的图，一个geom_***确定一种形状。更多几何形状推荐阅读这里
  • geom_bar()
  • geom_density()
  • geom_freqpoly()
  • geom_histogram()
  • geom_violin()
  • geom_boxplot()
  • geom_col()
  • geom_point()
  • geom_smooth()
  • geom_tile()
  • geom_density2d()
  • geom_bin2d()
  • geom_hex()
  • geom_count()
  • geom_text()
  • geom_sf()
  • stats: 统计变换
  • scales: 标度
  • coord: 坐标系统
  • facet: 分面
  • layer： 增加图层
  • theme: 主题风格
  • save: 保存图片
图片

开始

{BLOCK GGPLOT2-GEOM-10, TYPE="TRY"} 
R语言数据类型，有字符串型、数值型、因子型、逻辑型、日期型等。
ggplot2会将字符串型、因子型、逻辑型默认为**离散变量**，而数值型默认为**连续变量**，将日期时间为**日期变量**：

- **离散变量**: 字符串型<chr>, 因子型<fct>, 逻辑型<lgl>

- **连续变量**: 双精度数值<dbl>, 整数数值<int>

- **日期变量**: 日期<date>, 时间<time>, 日期时间<dttm>


我们在呈现数据的时候，可能会同时用到多种类型的数据，比如

* 一个离散
* 一个连续

* 两个离散
* 两个连续
* 一个离散, 一个连续

* 三个连续

导入数据

R 
gapdata <- read_csv("./demo_data/gapminder.csv")
gapdata

检查数据

是否有缺失值

R 
gapdata %>%
  summarise(
    across(everything(), ~ sum(is.na(.)))
  )
  • country 代表国家
  • countinet 表示所在的洲
  • year 时间
  • lifeExp 平均寿命
  • pop 人口数量
  • gdpPercap 人均GDP
{BLOCK GGPLOT2-GEOM-13, TYPE = "TRY"} 
接下来，我们需要思考我们应该选择什么样的图，呈现这些不同类型的数据，探索数据背后的故事

基本绘图

柱状图

常用于一个离散变量

R 
gapdata %>%
  ggplot(aes(x = continent)) +
  geom_bar()
R 
gapdata %>%
  ggplot(aes(x = reorder(continent, continent, length))) +
  geom_bar()
R 
gapdata %>%
  ggplot(aes(x = reorder(continent, continent, length))) +
  geom_bar() +
  coord_flip()
R 
# geom_bar vs stat_count
gapdata %>%
  ggplot(aes(x = continent)) +
  stat_count()
R 
gapdata %>% count(continent)

可见，geom_bar() 自动完成了这个统计，更多geom与stat对应关系见这里

R 
gapdata %>%
  distinct(continent, country) %>%
  ggplot(aes(x = continent)) +
  geom_bar()

我个人比较喜欢先统计，然后画图

R 
gapdata %>%
  distinct(continent, country) %>%
  group_by(continent) %>%
  summarise(num = n()) %>%
  ggplot(aes(x = continent, y = num)) +
  geom_col()

直方图

常用于一个连续变量

R 
gapdata %>%
  ggplot(aes(x = lifeExp)) +
  geom_histogram() # corresponding to stat_bin()
R 
gapdata %>%
  ggplot(aes(x = lifeExp)) +
  geom_histogram(binwidth = 1)

geom_histograms(), 默认使用 position = "stack"

R 
gapdata %>%
  ggplot(aes(x = lifeExp, fill = continent)) +
  geom_histogram()

也可以指定 position = "identity"

R 
gapdata %>%
  ggplot(aes(x = lifeExp, fill = continent)) +
  geom_histogram(position = "identity")

频次图

R 
gapdata %>%
  ggplot(aes(x = lifeExp, color = continent)) +
  geom_freqpoly()

密度图

R 
#' smooth histogram = density plot
gapdata %>%
  ggplot(aes(x = lifeExp)) +
  geom_density()

如果不喜欢下面那条线，可以这样

R 
gapdata %>%
  ggplot(aes(x = lifeExp)) +
  geom_line(stat = "density")

geom_density() 中adjust 用于调节bandwidth, adjust = 1/2 means use half of the default bandwidth.

R 
gapdata %>%
  ggplot(aes(x = lifeExp)) +
  geom_density(adjust = 1)

gapdata %>%
  ggplot(aes(x = lifeExp)) +
  geom_density(adjust = 0.2)
R 
gapdata %>%
  ggplot(aes(x = lifeExp, color = continent)) +
  geom_density()
R 
gapdata %>%
  ggplot(aes(x = lifeExp, fill = continent)) +
  geom_density(alpha = 0.2)
R 
gapdata %>%
  filter(continent != "Oceania") %>%
  ggplot(aes(x = lifeExp, fill = continent)) +
  geom_density(alpha = 0.2)

直方图和密度图画在一起。注意y = stat(density) 表示y是由x新生成的变量，这是一种固定写法，类似的还有stat(count), stat(level)

R 
gapdata %>%
  filter(continent != "Oceania") %>%
  ggplot(aes(x = lifeExp, y = stat(density))) +
  geom_histogram(aes(fill = continent)) +
  geom_density()

箱线图

一个离散变量 + 一个连续变量，思考下结果为什么是这样？

R 
gapdata %>%
  ggplot(aes(x = year, y = lifeExp)) +
  geom_boxplot()

数据框中的year变量是数值型，需要先转换成因子型，弄成离散型变量

R 
gapdata %>%
  ggplot(aes(x = as.factor(year), y = lifeExp)) +
  geom_boxplot()

明确指定分组变量

R 
gapdata %>%
  ggplot(aes(x = year, y = lifeExp)) +
  geom_boxplot(aes(group = year))
R 
gapdata %>%
  ggplot(aes(x = year, y = lifeExp)) +
  geom_violin(aes(group = year)) +
  geom_jitter(alpha = 1 / 4) +
  geom_smooth(se = FALSE)

抖散图

点重叠的处理方案

R 
gapdata %>% 
  ggplot(aes(x = continent, y = lifeExp)) +
  geom_point()
R 
gapdata %>% 
  ggplot(aes(x = continent, y = lifeExp)) +
  geom_jitter()
R 
gapdata %>% 
  ggplot(aes(x = continent, y = lifeExp)) +
  geom_boxplot()
R 
gapdata %>% 
  ggplot(aes(x = continent, y = lifeExp)) +
  geom_boxplot() +
  geom_jitter()
R 
gapdata %>%
  ggplot(aes(x = continent, y = lifeExp)) +
  geom_jitter() +
  stat_summary(fun.y = median, colour = "red", geom = "point", size = 5)
R 
gapdata %>%
  ggplot(aes(reorder(x = continent, lifeExp), y = lifeExp)) +
  geom_jitter() +
  stat_summary(fun.y = median, colour = "red", geom = "point", size = 5)

注意到我们已经提到过 stat_count / stat_bin / stat_summary

R 
gapdata %>%
  ggplot(aes(x = continent, y = lifeExp)) +
  geom_violin(
    trim = FALSE,
    alpha = 0.5
  ) +
  stat_summary(
    fun.y = mean,
    fun.ymax = function(x) {
      mean(x) + sd(x)
    },
    fun.ymin = function(x) {
      mean(x) - sd(x)
    },
    geom = "pointrange"
  )

山峦图

常用于一个离散变量 + 一个连续变量

R 
gapdata %>%
  ggplot(aes(
    x = lifeExp,
    y = continent,
    fill = continent
  )) +
  ggridges::geom_density_ridges()
R 
# https://learnui.design/tools/data-color-picker.html#palette
gapdata %>%
  ggplot(aes(
    x = lifeExp,
    y = continent,
    fill = continent
  )) +
  ggridges::geom_density_ridges() +
  scale_fill_manual(
    values = c("#003f5c", "#58508d", "#bc5090", "#ff6361", "#ffa600")
  )
R 
gapdata %>%
  ggplot(aes(
    x = lifeExp,
    y = continent,
    fill = continent
  )) +
  ggridges::geom_density_ridges() +
  scale_fill_manual(
    values = colorspace::sequential_hcl(5, palette = "Peach")
  )

散点图

常用于两个连续变量

R 
gapdata %>%
  ggplot(aes(x = gdpPercap, y = lifeExp)) +
  geom_point()
R 
gapdata %>%
  ggplot(aes(x = log(gdpPercap), y = lifeExp)) +
  geom_point()
R 
gapdata %>%
  ggplot(aes(x = gdpPercap, y = lifeExp)) +
  geom_point() +
  scale_x_log10() # A better way to log transform
R 
gapdata %>%
  ggplot(aes(x = gdpPercap, y = lifeExp)) +
  geom_point(aes(color = continent))
R 
gapdata %>%
  ggplot(aes(x = gdpPercap, y = lifeExp)) +
  geom_point(alpha = (1 / 3), size = 2)
R 
gapdata %>%
  ggplot(aes(x = gdpPercap, y = lifeExp)) +
  geom_point() +
  geom_smooth()
R 
gapdata %>%
  ggplot(aes(x = gdpPercap, y = lifeExp)) +
  geom_point() +
  geom_smooth(lwd = 3, se = FALSE)
R 
gapdata %>%
  ggplot(aes(x = gdpPercap, y = lifeExp)) +
  geom_point() +
  geom_smooth(lwd = 3, se = FALSE, method = "lm")
R 
gapdata %>%
  ggplot(aes(x = gdpPercap, y = lifeExp, color = continent)) +
  geom_point() +
  geom_smooth(lwd = 3, se = FALSE, method = "lm")
R 
jCountries <- c("Canada", "Rwanda", "Cambodia", "Mexico")

gapdata %>%
  filter(country %in% jCountries) %>%
  ggplot(aes(x = year, y = lifeExp, color = country)) +
  geom_line() +
  geom_point()
R 
gapdata %>%
  filter(country %in% jCountries) %>%
  ggplot(aes(
    x = year, y = lifeExp,
    color = reorder(country, -1 * lifeExp, max)
  )) +
  geom_line() +
  geom_point()

这是一种技巧，但我更推荐以下方法

R 
d1 <- gapdata %>%
  filter(country %in% jCountries) %>%
  group_by(country) %>%
  mutate(end_label = if_else(year == max(year), country, NA_character_))

d1
R 
d1 %>% ggplot(aes(
  x = year, y = lifeExp, color = country
)) +
  geom_line() +
  geom_point() +
  geom_label(aes(label = end_label)) +
  theme(legend.position = "none")

如果觉得麻烦，就用gghighlight宏包吧

R 
gapdata %>%
  filter(country %in% jCountries) %>%
  ggplot(aes(
    x = year, y = lifeExp, color = country
  )) +
  geom_line() +
  geom_point() +
  gghighlight::gghighlight()

点线图

R 
gapdata %>%
  filter(continent == "Asia" &amp; year == 2007) %>%
  ggplot(aes(x = lifeExp, y = country)) +
  geom_point()
R 
gapdata %>%
  filter(continent == "Asia" &amp; year == 2007) %>%
  ggplot(aes(
    x = lifeExp,
    y = reorder(country, lifeExp)
  )) +
  geom_point(color = "blue", size = 2) +
  geom_segment(aes(
    x = 40,
    xend = lifeExp,
    y = reorder(country, lifeExp),
    yend = reorder(country, lifeExp)
  ),
  color = "lightgrey"
  ) +
  labs(
    x = "Life Expectancy (years)",
    y = "",
    title = "Life Expectancy by Country",
    subtitle = "GapMinder data for Asia - 2007"
  ) +
  theme_minimal() +
  theme(
    panel.grid.major = element_blank(),
    panel.grid.minor = element_blank()
  )

分面

如果想分别画出每个洲的寿命分布图，我们想到的是这样

R 
gapdata %>% 
  filter(continent == "Africa") %>% 
  ggplot(aes(x = lifeExp)) +
  geom_density()

gapdata %>% 
  filter(continent == "Americas") %>% 
  ggplot(aes(x = lifeExp)) +
  geom_density()

gapdata %>% 
  filter(continent == "Asia") %>% 
  ggplot(aes(x = lifeExp)) +
  geom_density()

gapdata %>% 
  filter(continent == "Europe") %>% 
  ggplot(aes(x = lifeExp)) +
  geom_density()

gapdata %>% 
  filter(continent == "Oceania") %>% 
  ggplot(aes(x = lifeExp)) +
  geom_density()

事实上，ggplot2的分面，可以很快捷的完成。分面有两个

  • facet_grid()
  • facet_wrap()

facet_grid()

  • create a grid of graphs, by rows and columns
  • use vars() to call on the variables
  • adjust scales with scales = "free"
R 
gapdata %>%
  ggplot(aes(x = lifeExp)) +
  geom_density() +
  facet_grid(. ~ continent)
R 
gapdata %>%
  filter(continent != "Oceania") %>%
  ggplot(aes(x = lifeExp, fill = continent)) +
  geom_histogram() +
  facet_grid(continent ~ .)
R 
gapdata %>%
  filter(continent != "Oceania") %>%
  ggplot(aes(x = lifeExp, y = stat(density))) +
  geom_histogram(aes(fill = continent)) +
  geom_density() +
  facet_grid(continent ~ .)

facet_wrap()

  • create small multiples by "wrapping" a series of plots
  • use vars() to call on the variables
  • nrow and ncol arguments for dictating shape of grid
R 
gapdata %>%
  ggplot(aes(x = gdpPercap, y = lifeExp, color = continent)) +
  geom_point(show.legend = FALSE) +
  facet_wrap(~continent)

文本标注

R 
gapdata %>%
  ggplot(aes(x = gdpPercap, y = lifeExp)) +
  geom_point() +
  ggforce::geom_mark_ellipse(aes(
    filter = gdpPercap &gt; 70000,
    label = "Rich country",
    description = "What country are they?"
  ))
R 
ten_countries <- gapdata %>%
  distinct(country) %>%
  pull() %>%
  sample(10)
R 
library(ggrepel)
gapdata %>%
  filter(year == 2007) %>%
  mutate(
    label = ifelse(country %in% ten_countries, as.character(country), "")
  ) %>%
  ggplot(aes(log(gdpPercap), lifeExp)) +
  geom_point(
    size = 3.5,
    alpha = .9,
    shape = 21,
    col = "white",
    fill = "#0162B2"
  ) +
  geom_text_repel(
    aes(label = label),
    size = 4.5,
    point.padding = .2,
    box.padding = .3,
    force = 1,
    min.segment.length = 0
  ) +
  theme_minimal(14) +
  theme(
    legend.position = "none",
    panel.grid.minor = element_blank()
  ) +
  labs(
    x = "log(GDP per capita)",
    y = "life expectancy"
  )

errorbar图

R 
avg_gapdata <- gapdata %>%
  group_by(continent) %>%
  summarise(
    mean = mean(lifeExp),
    sd = sd(lifeExp)
  )
avg_gapdata
R 
avg_gapdata %>%
  ggplot(aes(continent, mean, fill = continent)) +
  # geom_col(alpha = 0.5) +
  geom_point() +
  geom_errorbar(aes(ymin = mean - sd, ymax = mean + sd), width = 0.25)

椭圆图

R 
gapdata %>%
  ggplot(aes(x = log(gdpPercap), y = lifeExp)) +
  geom_point() +
  stat_ellipse(type = "norm", level = 0.95)

2D 密度图

与一维的情形geom_density()类似， geom_density_2d(), geom_bin2d(), geom_hex()常用于刻画两个变量构成的二维区间的密度

R 
gapdata %>%
  ggplot(aes(x = gdpPercap, y = lifeExp)) +
  geom_bin2d()
R 
gapdata %>%
  ggplot(aes(x = gdpPercap, y = lifeExp)) +
  geom_hex()

马赛克图

geom_tile()， geom_contour()， geom_raster()常用于3个变量

R 
gapdata %>%
  group_by(continent, year) %>%
  summarise(mean_lifeExp = mean(lifeExp)) %>%
  ggplot(aes(x = year, y = continent, fill = mean_lifeExp)) +
  geom_tile() +
  scale_fill_viridis_c()

事实上可以有更好的呈现方式

R 
gapdata %>%
  group_by(continent, year) %>%
  summarise(mean_lifeExp = mean(lifeExp)) %>%
  ggplot(aes(x = year, y = continent, size = mean_lifeExp)) +
  geom_point()
R 
gapdata %>%
  group_by(continent, year) %>%
  summarise(mean_lifeExp = mean(lifeExp)) %>%
  ggplot(aes(x = year, y = continent, size = mean_lifeExp)) +
  geom_point(shape = 21, color = "red", fill = "white") +
  scale_size_continuous(range = c(7, 15)) +
  geom_text(aes(label = round(mean_lifeExp, 2)), size = 3, color = "black") +
  theme(legend.position = "none")

课后思考题

哪画图的代码中，哪两张图的结果是一样？为什么？

R 
library(tidyverse)
tbl <-
  tibble(
    x = rep(c(1, 2, 3), times = 2),
    y = 1:6,
    group = rep(c("group1", "group2"), each = 3)
  )
ggplot(tbl, aes(x, y)) + geom_line()
ggplot(tbl, aes(x, y, group = group)) + geom_line()
ggplot(tbl, aes(x, y, fill = group)) + geom_line()
ggplot(tbl, aes(x, y, color = group)) + geom_line()

参考资料

R 
# remove the objects
# rm(list=ls())
rm(avg_gapdata, d1, df, gapdata, jCountries, ten_countries)
R 
pacman::p_unload(pacman::p_loaded(), character.only = TRUE)