# 饼图中的南辕北辙

```
# install needed R packages
::update_packages(c('magrittr', 'tidyverse', 'cowplot'), upgrade = TRUE) remotes
```

# 1 Beginning

为了展示毕业论文的数据，我想要画个饼图。`pie()`

虽然简单，但我肯定要用高大上的“哥哥作图”（**ggplot2**）。

```
library(magrittr)
library(ggplot2)
<- tibble::tribble(
df ~type, ~n,
'GDS', 1772,
'other GSE', 18309
%>% dplyr::mutate(percent = n/sum(n), label = scales::percent(percent))
)
df## # A tibble: 2 x 4
## type n percent label
## <chr> <dbl> <dbl> <chr>
## 1 GDS 1772 0.0882 9%
## 2 other GSE 18309 0.912 91%
```

# 2 Development

Althought I already know pie chart is a variant of bar plot with polar coordinate, I don’t know the precise code. As usual, Stack Overflow provides a good example.

I applied the code, but the order of label seems to be wrong:

```
ggplot(df) +
geom_bar(aes(x = "", y = percent, fill = type), stat="identity", width = 1) +
coord_polar("y", start = 0) +
theme_void() +
geom_text(aes(x = 1, y = cumsum(percent) - percent/2, label = label))
```

After a while, I find a solution:

```
ggplot(df) +
geom_bar(aes(x = "", y = percent, fill = factor(type, levels = rev(type))), stat="identity", width = 1) +
coord_polar("y", start = 0)+
theme_void()+
geom_text(aes(x = 1, y = cumsum(percent) - percent/2, label = label))
```

# 3 Climax

However, I still don’t understand what happens, so I start to explore it step by step.

I refered to the official documentation for how pie chart is evolved from bar plot.

To understand why the order of label is incorrect, we need to examine what happens in the original bar plot:

```
::plot_grid(
cowplotggplot(df) +
geom_col(aes(type, percent, fill = type)) +
geom_text(aes(x=type, y = percent/2, label=type)),
ggplot(df) +
geom_col(aes('', percent, fill = type)) +
geom_text(aes(x='', y = cumsum(percent) - percent/2, label=type)),
ggplot(dplyr::arrange(df, dplyr::desc(type)))+
geom_col(aes('', percent, fill = type)) +
geom_text(aes(x='', y = cumsum(percent) - percent/2, label=type)),
nrow = 1
)
```

- y of label is determined by the row order in data, i,e, the earlier a row appears, the smaller y value, and the lower position.
- The order of bar is determined by factor level order (default is lexicographic order of level strings). The smaller level order the higher position.

So `dplyr::arrange(..., dplyr::desc(...))`

can make top row have small y value and big level order, thus sovle the problem ^{1}.

# 4 Afterword

When I come back to polish this post on 2019/03/03, I ponder the official documentation for half an hour and realize I still haven’t fully understand the transition from bar plot to pie chart.

I think the documentation is not quiet clear. The most confusing thing is how radius is determined, since it’s neither mentioned nor deducible from examples.

The examples have two drawbacks:

- the yuse real data, so reader can’t know the exact value of y.
- the corresponding bar plot is not shown, so reader have to imagine it.

Therefore, I write the following examples, hoping they can save time when I forget how pie chart is drawn the next time:

```
<- tibble::tribble(
tbl ~class, ~value,
'A', 1,
'B', 2,
'C', 4
)
<- ggplot(tbl) + geom_col(aes(class, value), width = 1)
bar ::plot_grid(
cowplot
bar,+ coord_polar(theta = 'x'),
bar + coord_polar(theta = 'y'),
bar nrow = 1
)
```

From the aboving plots, we can know that:

`theta = 'x'`

: x becomes angle, y becomes radius`theta = 'y'`

: x becomes radius, y becomes angle

```
<- ggplot(tbl) + geom_col(aes('1', value, fill = class), width = 1)
stack ::plot_grid(
cowplot
stack,+ coord_polar(theta = 'x'),
stack + coord_polar(theta = 'y'),
stack nrow = 1
)
```

In the aboving plots, `fill`

divide y value:

- When y is radius, we have concentric circles.
- When y is angle, we have sectors.

`factor(type, levels = rev(type))`

is a coincidence, it only works in this case.↩︎