Coloring boxplot outlier points in ggplot2?

2020-01-31 02:08发布

站内文章 / 前沿技术

43 0

放荡不羁爱自由

女 | 书童

私信

可以将文章内容翻译成中文,广告屏蔽插件可能会导致该功能失效(如失效，请关闭广告屏蔽插件后再试):

问题:

How can I color the outlier points in ggplot2? I want them to be the same color as the boxplot itself. colour= is not enough to do this.

Example:

p <- ggplot(mtcars, aes(factor(cyl), mpg))
p + geom_boxplot(aes(colour=factor(cyl)))

I want to color the outliers by factor(cyl) as well. This does not work:

> p <- ggplot(mtcars, aes(factor(cyl), mpg))
> p + geom_boxplot(aes(colour=factor(cyl), outlier.colour=factor(cyl)))

回答1:

In order to color the outlier points the same as your boxplots, you're going to need to calculate the outliers and plot them separately. As far as I know, the built-in option for coloring outliers colors all outliers the same color.

The help file example

Using the same data as the 'geom_boxplot' help file:

ggplot(mtcars, aes(x=factor(cyl), y=mpg, col=factor(cyl))) +
    geom_boxplot()

Coloring the outlier points

Now there may be a more streamlined way to do this, but I prefer to calculate things by hand, so I don't have to guess what's going on under the hood. Using the 'plyr' package, we can quickly get the upper and lower limits for using the default (Tukey) method for determining an outlier, which is any point outside the range [Q1 - 1.5 * IQR, Q3 + 1.5 * IQR]. Q1 and Q3 are the 1/4 and 3/4 quantiles of the data, and IQR = Q3 - Q1. We could write this all as one huge statement, but since the 'plyr' package's 'mutate' function will allow us to reference newly-created columns, we might as well split it up for easier reading/debugging, like so:

library(plyr)
plot_Data <- ddply(mtcars, .(cyl), mutate, Q1=quantile(mpg, 1/4), Q3=quantile(mpg, 3/4), IQR=Q3-Q1, upper.limit=Q3+1.5*IQR, lower.limit=Q1-1.5*IQR)

We use the 'ddply' function, because we are inputting a data frame and wanting a data frame as output ("d->d" ply). The 'mutate' function in the above 'ddply' statement is preserving the original data frame and adding additional columns, and the specification of .(cyl) is telling the functions to be calculated for each grouping of 'cyl' values.

At this point, we can now plot the boxplot and then overwrite the outliers with new, colored points.

ggplot() +
    geom_boxplot(data=plot_Data, aes(x=factor(cyl), y=mpg, col=factor(cyl))) + 
    geom_point(data=plot_Data[plot_Data$mpg > plot_Data$upper.limit | plot_Data$mpg < plot_Data$lower.limit,], aes(x=factor(cyl), y=mpg, col=factor(cyl)))

What we are doing in the code is to specify an empty 'ggplot' layer and then adding the boxplot and point geometries using independent data. The boxplot geometry could use the original data frame, but I am using our new 'plot_Data' to be consistent. The point geometry is then only plotting the outlier points, using our new 'lower.limit' and 'upper.limit' columns to determine outlier status. Since we use the same specification for the 'x' and 'col' aesthetic arguments, the colors are magically matched between the boxplots and the corresponding outlier points.

Update: The OP requested a more complete explanation of the 'ddply' function used in this code. Here it is:

The 'plyr' family of functions are basically a way of subsetting data and performing a function on each subset of the data. In this particular case, we have the statement:

ddply(mtcars, .(cyl), mutate, Q1=quantile(mpg, 1/4), Q3=quantile(mpg, 3/4), IQR=Q3-Q1, upper.limit=Q3+1.5*IQR, lower.limit=Q1-1.5*IQR)

Let's break this down in the order the statement would be written. First, the selection of the 'ddply' function. We want to calculate the lower and upper limits for each value of 'cyl' in the 'mtcars' data. We could write a 'for' loop or other statement to calculate these values, but then we would have to write another logic block later to assess outlier status. Instead, we want to use 'ddply' to calculate the lower and upper limits and add those values to every line. We choose 'ddply' (as opposed to 'dlply', 'd_ply', etc.), because we are inputting a data frame and wanting a data frame as output. This gives us:

ddply(

We want to perform the statement on the 'mtcars' data frame, so we add that.

ddply(mtcars,

Now, we want to perform our calculations using the 'cyl' values as a grouping variable. We use the 'plyr' function .() to refer to the variable itself rather than to the variable's value, like so:

ddply(mtcars, .(cyl),

The next argument specifies the function to apply to every group. We want our calculation to add new rows to the old data, so we choose the 'mutate' function. This preserves the old data and adds the new calculations as new columns. This is in contrast to other functions like 'summarize', which removes all of the old columns except the grouping varaible(s).

ddply(mtcars, .(cyl), mutate,

The final series of arguments are all of the new columns of data we want to create. We define these by specifying a name (unquoted) and an expression. First, we create the 'Q1' column.

ddply(mtcars, .(cyl), mutate, Q1=quantile(mpg, 1/4),

The 'Q3' column is calculated similarly.

ddply(mtcars, .(cyl), mutate, Q1=quantile(mpg, 1/4), Q3=quantile(mpg, 3/4),

Luckily, with the 'mutate' function, we can use newly created columns as part of the definition of other columns. This saves us from having to write one giant function or from having to run multiple functions. We need to use 'Q1' and 'Q3' in the calculation of the inter-quartile range for the 'IQR' variable, and that's easy with the 'mutate' function.

ddply(mtcars, .(cyl), mutate, Q1=quantile(mpg, 1/4), Q3=quantile(mpg, 3/4), IQR=Q3-Q1,

We're finally where we want to be now. We technically don't need the 'Q1', 'Q3', and 'IQR' columns, but it does make our lower limit and upper limit equations a lot easier to read and debug. We can write our expression just like the theoretical formula: limits=+/- 1.5 * IQR

ddply(mtcars, .(cyl), mutate, Q1=quantile(mpg, 1/4), Q3=quantile(mpg, 3/4), IQR=Q3-Q1, upper.limit=Q3+1.5*IQR, lower.limit=Q1-1.5*IQR)

Cutting out the middle columns for readability, this is what the new data frame looks like:

plot_Data[, c(-3:-11)]
#     mpg cyl    Q1    Q3  IQR upper.limit lower.limit
# 1  22.8   4 22.80 30.40 7.60      41.800      11.400
# 2  24.4   4 22.80 30.40 7.60      41.800      11.400
# 3  22.8   4 22.80 30.40 7.60      41.800      11.400
# 4  32.4   4 22.80 30.40 7.60      41.800      11.400
# 5  30.4   4 22.80 30.40 7.60      41.800      11.400
# 6  33.9   4 22.80 30.40 7.60      41.800      11.400
# 7  21.5   4 22.80 30.40 7.60      41.800      11.400
# 8  27.3   4 22.80 30.40 7.60      41.800      11.400
# 9  26.0   4 22.80 30.40 7.60      41.800      11.400
# 10 30.4   4 22.80 30.40 7.60      41.800      11.400
# 11 21.4   4 22.80 30.40 7.60      41.800      11.400
# 12 21.0   6 18.65 21.00 2.35      24.525      15.125
# 13 21.0   6 18.65 21.00 2.35      24.525      15.125
# 14 21.4   6 18.65 21.00 2.35      24.525      15.125
# 15 18.1   6 18.65 21.00 2.35      24.525      15.125
# 16 19.2   6 18.65 21.00 2.35      24.525      15.125
# 17 17.8   6 18.65 21.00 2.35      24.525      15.125
# 18 19.7   6 18.65 21.00 2.35      24.525      15.125
# 19 18.7   8 14.40 16.25 1.85      19.025      11.625
# 20 14.3   8 14.40 16.25 1.85      19.025      11.625
# 21 16.4   8 14.40 16.25 1.85      19.025      11.625
# 22 17.3   8 14.40 16.25 1.85      19.025      11.625
# 23 15.2   8 14.40 16.25 1.85      19.025      11.625
# 24 10.4   8 14.40 16.25 1.85      19.025      11.625
# 25 10.4   8 14.40 16.25 1.85      19.025      11.625
# 26 14.7   8 14.40 16.25 1.85      19.025      11.625
# 27 15.5   8 14.40 16.25 1.85      19.025      11.625
# 28 15.2   8 14.40 16.25 1.85      19.025      11.625
# 29 13.3   8 14.40 16.25 1.85      19.025      11.625
# 30 19.2   8 14.40 16.25 1.85      19.025      11.625
# 31 15.8   8 14.40 16.25 1.85      19.025      11.625
# 32 15.0   8 14.40 16.25 1.85      19.025      11.625

Just to give a contrast, if we were to do the same 'ddply' statement with the 'summarize' function, instead, we would have all of the same answers but without the columns of the other data.

ddply(mtcars, .(cyl), summarize, Q1=quantile(mpg, 1/4), Q3=quantile(mpg, 3/4), IQR=Q3-Q1, upper.limit=Q3+1.5*IQR, lower.limit=Q1-1.5*IQR)
#   cyl    Q1    Q3  IQR upper.limit lower.limit
# 1   4 22.80 30.40 7.60      41.800      11.400
# 2   6 18.65 21.00 2.35      24.525      15.125
# 3   8 14.40 16.25 1.85      19.025      11.625

回答2:

Update (2015-03-31): see @tarch's solution for ggplot2 >= 1.0.0

solution for ggplot2 <= 0.9.3 is below.

As @koshke said, having the outliers colored like the lines of the box (not the fill color) is now easily possible by setting outlier.colour = NULL:

p <- ggplot(mtcars, aes(x=factor(cyl), y=mpg, col=factor(cyl)))
p + geom_boxplot(outlier.colour = NULL)

outlier.colour must be written with "ou"
outlier.colour must be outside aes ()

I'm posting this as a late answer because I find myself looking this up again and again and I posted it also for the related question Boxplot, how to match outliers' color to fill aesthetics?

回答3:

I found a solution to the fact that setting geom_boxplot(outlier.colour = NULL) doesn't work anymore in newest versions of R (@hamy speaks about version 1.0.0 of ggplot2).

In order to replicate the behaviour that @cbeleites proposed you simply need to use the following code:

update_geom_defaults("point", list(colour = NULL))
m <- ggplot(movies, aes(y = votes, x = factor(round(rating)),
            colour = factor(Animation)))
m + geom_boxplot() + scale_y_log10()

as expected this produces plot with points that match the line color.

Of course one should remember to restore the default if he needs to draw multiple plots:

update_geom_defaults("point", list(colour = "black"))

The solution was found by reading the ggplot2 changelog on github:

The outliers of geom_boxplot() use the default colour, size and shape from geom_point(). Changing the defaults of geom_point() with update_geom_defaults() will apply the same changes to the outliers of geom_boxplot(). Changing the defaults for the outliers was previously not possible. (@ThierryO, #757)

Posted here as well: ggplot2 boxplot, how do i match the outliers' color to fill aesthetics?

回答4:

If there is a need to change shape or color of the outlier points according to different factor (not the same which is used for making boxplot groups) then answer of @Dinre can be adapted.

Color of points can be changed only if the color isn't used for boxplot themselves (you can't use two variables for colors).

Using the data plot_Data and code from the @Dinre answer - color of outliers depend on factor carb. By adding argument outlier.shape = NA to geom_boxplot() original outliers are removed to ensure that they are not over-plotted by geom_point().

ggplot() +
  geom_boxplot(data=plot_Data, aes(x=factor(cyl), y=mpg),outlier.shape = NA) + 
  geom_point(data=plot_Data[plot_Data$mpg > plot_Data$upper.limit | 
                              plot_Data$mpg < plot_Data$lower.limit,], 
             aes(x=factor(cyl), y=mpg, color=factor(carb)))

To change the shape of points:

ggplot() +
  geom_boxplot(data=plot_Data, aes(x=factor(cyl), y=mpg),outlier.shape = NA) + 
  geom_point(data=plot_Data[plot_Data$mpg > plot_Data$upper.limit | 
                              plot_Data$mpg < plot_Data$lower.limit,], 
             aes(x=factor(cyl), y=mpg, shape=factor(carb)))

回答5:

The outliers automatically inherits the colours from the box again in ggplot2 * 1.0.1.9003.

https://github.com/hadley/ggplot2/issues/1400

```{r}
library(ggplot2)
point_size=10
ggplot(mtcars, aes(x=factor(cyl), y=mpg, col=factor(cyl))) +
geom_boxplot(outlier.size = point_size)
```

Boxplot

回答6:

Just in case you really want to keep the boxes black:

Another possible solution is to

store in a variable whether a point is an outlier,
suppress outlier plotting with outlier.color = NA, and
plot only those points that are outliers and use the col aesthetic in this call to geom_point()

Step 1: Define a function to determine whether a point is an outlier:

is.outlier <- function (x) {
  x < quantile(x, .25) - 1.5 * IQR(x) |
    x > quantile(x, .75) + 1.5 * IQR(x)
}

Step 2: Group data and calculate outliers with this function for the groups

diamonds %>% group_by(cut) %>%
  mutate(outlier.p = is.outlier(price)) %>%
  ungroup() -> diamonds

Step 3: Create plot

ggplot(diamonds, aes(x = cut, y = price, fill = cut)) +
  geom_boxplot(outlier.color = NA) +
  geom_point(data = diamonds[diamonds$outlier.p,], aes(col = cut))

Please note that you have to subset your dataset in the geom_point() call (last line) lest all points are plotted.