The examples below use the downloads dataset, which also was used for the presentation Working with data in R. This dataset is available as an Excel file, which we will import using the readxl package. Furthermore, ggplot2 is part of the tidyverse package, so we also load that:
library(readxl)library(tidyverse)
── Attaching core tidyverse packages ──────────────────────── tidyverse 2.0.0 ──
✔ dplyr 1.2.0 ✔ readr 2.1.6
✔ forcats 1.0.1 ✔ stringr 1.6.0
✔ ggplot2 4.0.2 ✔ tibble 3.3.1
✔ lubridate 1.9.5 ✔ tidyr 1.3.2
✔ purrr 1.2.1
── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
✖ dplyr::filter() masks stats::filter()
✖ dplyr::lag() masks stats::lag()
ℹ Use the conflicted package (<http://conflicted.r-lib.org/>) to force all conflicts to become errors
We read the downloads dataset from the Excel file, and assign it to a tibble named downloads. Moreover, we only keep the observations, where the size variable is strictly larger than zero:
We manually make a Table-of-Variables with the meta-information about the variables in the dataset:
Variable
Description
Range
Usage
machineName
Name the server
categorical (5 levels)
explanatory
userID
Id for user
categorical (35811 levels)
not used
size
download size in bytes
numerical (3 to 14518894)
response
time
download time in seconds
numerical (0.0437 to 1878.0762)
response
date
download date
date (Nov 22, 1994 to May 18, 1995)
explanatory
month
month of the date
categorical (1994-11 to 1995-05)
explanatory
ggplot2: The basic concepts
A ggplot-object is a syntactical description of a plot. You may think of it as a recipe in a cookbook. To actually cook the dish, that is to make the plot, you print the ggplot-object. By default the results of all R commands executed in the Console are automatically printed. Thus, as a result the plot will be generated on the computer screen. If you execute an R script by sourcing, then you might need to print explicitly using the print() command. Alternatively, you can use the ggsave() command to print the syntactical plot description into a graphics file to be used in scientific papers and/or presentations.
To write down a recipe for a dish you usually start with a blank sheet of paper. We do the same for our graphical recipes. The equivalent of a blank sheet of paper is generated by the command ggplot(). The ingredients for our plot is a dataset, which should be available as a data.frame or as a tibble. We should also specify what the ingredients should be used for. In the language of ggplot2 this is done by specifying aesthetics via the aes()command.
Suppose we want to use data from the tibble downloads, and that machineName should be on the x-axis and size on the y-axis. Then we write
ggplot(downloads,aes(x=machineName,y=size))
The reason that we don’t see any points, lines or the like is, of course, that we did not yet ask for such things to be made! Geometrical objects like points and lines are called geoms in ggplot2. But although we did not yet add any geoms to our plot, we see that the plot already recognized the range (and type) of the variables specified in the aesthetics.
A simple bar chart
To make a bar chart we add geom_col() to the syntactical description. In the code below we have also rescaled the size of the downloaded files to be measured in mega bytes instead of bytes. This is done by downscaling the size variables by a factor 1,000,000.
ggplot(downloads, aes(x = machineName, y = size/10^6)) +geom_col()
We notice, that machineName used on the x-axis is a categorical variable. In R the levels of a categorical variable by default are ordered alphabetically, which is also what we see on this plot. The y-axis shows the total number of mega bytes for each machine, i.e., the sum over all observations from each machine. The sum is computed automatically, but we could also have chosen to make those computations “manually” and made the plot like this instead:
Then it is also obvious how to change the code if we wanted the average download size, say, rather than the total download on the y-axis (try it yourself).
Flipping the bar chart
In the next R chunk we for the first time will try to save the syntactical description of a plot in an R variable. The benefit of doing this is that the syntactical description easily may be reused, possibly with variations. In the metaphor of a recipe in a cookbook think of giving a piece of paper with the recipe of your favorite dish to a friend. Then your friend may cook your dish, and possibly also with variations and new additions. Let’s try out this idea, and write down the recipe for the bar chart in an R variable called p:
p <-ggplot(downloads, aes(x = machineName, y = size/10^6)) +geom_col()
This does not yet produce a new plot. But a variable called p has appeared in the global environment, and if we gave the command p, then the graph would be plotted again. Now imagine you give the recipe to your friend. Your friend is happy and thank you for the wonderful recipe, but decide to cook the bar chart with horizontal bars instead of vertical bars. This is done by flipping the coordinate axes:
p +coord_flip()
Adding monthly download info
We can extend the graphics further by adding new aesthetics and/or geoms. Looking at the help page ?geom_col we see that the fill-aesthetic will be interpreted by geom_col(). To see the effect of this aesthetic on that geom we simply try it out.
p +aes(fill = month)
We see that the bars have the same height as before, and still visualize the total download size. But now the total download size also has been subdivided according to month. This is visualized by colors, and a legend for the interpretation of the colors is automatically added in the right panel of the plot.
Some other bar chart options
Above we realized that setting the fill-aesthetic results in a subdivision of the contributions to the bars. How this is displayed may be changed by setting the position-option in the geom_col() call. Let’s try it out!
p <-ggplot(downloads, aes(x = machineName, y = size/10^6, fill = month))p +geom_col(position ="dodge") ## Left/first plot
p +geom_col(position ="fill") ## Right/second plot
Caveat: Take a closer look at the output delivered via position = "dodge". Is it doing what you want? Or what you expected it to do? My guess is that it isn’t, and that you presumably wanted and expected the output from the following:
`summarise()` has regrouped the output.
ℹ Summaries were computed grouped by machineName and month.
ℹ Output is grouped by machineName.
ℹ Use `summarise(.groups = "drop_last")` to silence this message.
ℹ Use `summarise(.by = c(machineName, month))` for per-operation grouping
(`?dplyr::dplyr_by`) instead.
Very tricky question: Can you explain what is doing on here? (You might want to ask the teacher about this!)
A bar chart with ordered bars
Suppose we like the machines in the bar chart to be ordered according to increasing download size. One way to achieve this is to recode the variable machineName as a factor (in R categorical variables are called factors) with levels ordered according to increasing download size. Using the techniques presented in Working with data in R we generate a tibble containing the total download size from the five different machines, which we thereafter ordered according to total download size:
Finally, we can make the plot using the same ggplot() code as above. Thus, the same ggplot() code with a changed dataset (remember, that we made a new ordering of the levels of the variable machineName) will give a new plot:
ggplot(downloads, aes(x = machineName, y = size/10^6)) +geom_col()
Daily summary statistics
Next we want to visualize the number and the total size of the downloads done each date for each of the 6 machines. For later usage we also compute the cumulated number of downloads within each of the 6 machines over the dates. We do this via the following steps:
Using group_by() we group the dataset by both machineName and date.
Using summarize() we count the number and the total size of the downloads for within each machine and date.
Using mutate() we cumulate the number of downloads over the dates within the machines. We remark that cumsum makes the cumulative sum over the innermost grouping variable, which is date.
`summarise()` has regrouped the output.
ℹ Summaries were computed grouped by machineName and date.
ℹ Output is grouped by machineName.
ℹ Use `summarise(.groups = "drop_last")` to silence this message.
ℹ Use `summarise(.by = c(machineName, date))` for per-operation grouping
(`?dplyr::dplyr_by`) instead.
To make a scatter plot we use geom_point(). We save the ggplot-description in the variable p, so that it is easy to try out different layout features on the same plot. Please note that this, of course, will overwrite the previous content of p (which happened to be the description of the bar charts). Thus, after executing the following R chunk, p will contain the description of a scatter plot.
p <-ggplot(daily_downloads, aes(x = date, y = dl_count)) +geom_point()p
Plotting on the log-scale
To change the y-axis to be logarithmic we add scale_y_log10(). Please note, that for visualizations we often prefer the base-10 logarithm (whereas statisticians often use the natural logarithm for modeling). However, for some applications the natural logarithm or the base-2 logarithm might be the preferable choice.
p <- p +scale_y_log10()p
Points colored by machine
Remember, that p presently encodes a scatterplot, which is made using geom_point(). To color the points according to machineName we simply add this as an aesthetic.
p +aes(color = machineName)
Points shaped by machine
Alternatively, we can visualize the five different machines by different plotting symbold. This is done by adding a shape aesthetic instead.
p +aes(shape = machineName)
Please note that ggplot2 only contains six different plotting symbols. If you use the shape aesthetic on a categorical variable with more than 6 levels, then you get the following error message:
> p +aes(shape =factor(size_mb))Warning messages:1: The shape palette can deal with a maximum of 6 discrete values because more than 6 becomes difficult todiscriminate; you have 337. Consider specifying shapes manually if you must have them. 2: Removed 331 rows containing missing values (geom_point).
Bubble plot
The bubble plot is an extension of the classical scatter plot. Recall, that a scatter plot displays two numerical variables via the x-axis and the y-axis. The idea of the bubble plot is to visualize a third numerical variable by letting it encode the size of the points. The best human perception of size is achieved when the area (and not the diameter, say) of the points is taken to be proportional to this third numerical variable. This is exactly what is achieved by the size aesthetic:
p +aes(size = size_mb)
Please note, that “size” appearing on the left hand side is the name of the aesthetic known by ggplot2, whereas “size_mb” on the right hand side is the name of the variable inside the tibble downloads. It does not come as a surprise that the total download size is often large (large points) when the number of downloaded packages is large (large value on y-axis).
Points colored by download size
The ideas exemplified above may be combined. E.g. to make a stronger visualization of the daily total download size we may choose to make a combined usage of the size and the color aesthetic.
p +aes(size = size_mb, color = size_mb >2)
From the inserted legend in the right panel of the plot it should be clear how the colors are to be interpreted.
Faceting
Faceting is a clever way of visualizing additional categorical variables (one, two, or even more of them) by dividing the plot into several panels (either along the rows, the columns, or both the row and the columns). Here is an example:
p +facet_wrap(~machineName)
And here is another one:
p +facet_grid(. ~ machineName)
Cumulated total download size over the dates within machines
The geom called geom_line() is used to insert lines. In the code below you see a first attempt to write R code that visualizes the cumulated total download size over the dates.
ggplot(daily_downloads, aes(x = date, y = total_dl_count)) +geom_line()
However, in the code above we forgot to make the lines within the machines. This may be achieved by adding a group aesthetic as shown in the following code.
ggplot(daily_downloads, aes(x = date, y = total_dl_count)) +geom_line(aes(group = machineName))
A box plot
Let’s also try to make a boxplot…
p <-ggplot(daily_downloads, aes(x = machineName, y = size_mb)) +geom_boxplot()p
But perhaps the boxplot is more informative on the log-scale? Let’s try it out!
p +scale_y_log10()
An alternative to the classical boxplot is the so-called violin plot, which show the full probability distribution:
ggplot(daily_downloads, aes(x = machineName, y = size_mb)) +geom_violin() +scale_y_log10()
Saving plots
ggplots easily can be printed to a graphical device. The following code saves the most recently produced plot to a pdf-file called violin.pdf. Please note that ggsave() guesses the graphics format from the file name:
ggsave("violin.pdf")
Saving 7 x 5 in image
If a plot has been saved as an object, like p above, then it can be saved with ggsave() even if is not printed on the screen. Moreover, the size of the image can be changed if needed, as this command shows:
ggsave("p-plot.pdf", p, width=10, height=5)
Conclusion
The ggplot2 package is a powerful and versatile tool for making plots.
We think, that the generated plots are beautiful.
As far as we know some plots, e.g. using faceting, in practice are only producible in ggplot2.
Learning the syntax needed for making specific plots is a challenge. The best way (the only way!?) to learn is to practice. You may start by solving the exercise sheet. After you have gotten used to the basic ideas you can find a lot of help on the internet. We remark, that there have been several updates of the ggplot2 package over the recent years, and some of the old entries that pop up when you google a ggplot2-issue might be outdated.
Not all things can be made in ggplot2! For a geometrical object to be available it need to have a syntactical description, and it needs to be implemented. Other things require computer-hacks to be made (e.g. using different orderings of categorical variables in faceted plots).
Many geoms make statistical computations before the actual plot is made. This can be very practical. However, occasionally the computation done isn’t want you perhaps wanted. As e.g. was the case with geom_col(position = "dodge"). When in doubt, it can be more safe to do the needed computations manually (i.e. via tidyverse and dplyr).
