How to Speak ggplot2 Like a Native
Harlan D. Harris, PhD
October 16, 2013
Background
ggplot's philosophy
graphics as grammar
The computer translates it to an image:
spec + data.frame =
As we'll see later, a strength of ggplot is that you can do certain types of computations on your data from within the system. Here, it's computing a ratio of two things. In general, this power allows you to quickly explore your data, and different ways of presenting your data, without having to make changes to your data table.
advantages
ggplot2
Gripes
representations
illegibility
abuses R
introductions create misconceptions
- Hadley's book is a particularly bad example of this
Let's Make A Graph! (slowly)
goal
data
ggplot likes "long" data
simplest plot
bb_long_al <- subset(bb_long, lg=='AL')
p <- ggplot(data=bb_long_al,
mapping=aes(x=year, y=value,
color=stat)) +
layer(geom='line')
print(p)
p – building an object, not plotting on the screen ggplot – create a “core” ggplot object – required data – ggplot wants a data.frame mapping – all ggplots have an underlying x and y, and the mapping parameter describes how to map the underlying variables to the variables in a df x and y – realized as horiz and vert here, but might not be in other graphs. obvious mappings. Note that ggplot grabs the df col. names as default axis labels color – the color of the lines is mapped to the value of the Condition column. (the actual colors are set by default here – will override later)
- - somewhat strangely, ggplot objects are built piece by piece using addition as the operator. Sometimes you're adding to an existing piece, sometimes to a new piece. layer – things that are drawn are in layers. Layers have geoms, which are geometric shapes like points, lines, bars, etc.
structure (not on the test!)
add layers
structure so far
scales
Mappings of data to representations of displayable attributes.
p3 <- p2 + scale_y_continuous("Total", limits=c(0,15000),
breaks=c(0,5000,10000,12000)) +
scale_x_continuous("Year") +
scale_color_manual("Stat",
values=c('darkgreen','darkblue','darkred'))
coordinates and scales
themes
Adjust properties of a graph as a whole.
p4 <- p3 + ggtitle("Baseball!") +
theme_bw() +
theme(plot.title=element_text(color='#FF8888', size=40))
Note that although this is last in series of function calls added together, it's actually modifying the base ggplot object, not any of the layers. This is unfortunate. Not going into themes, but basically you can modify lots of the underlying lines and boxes and colors that make up a ggplot plot, either on a case-by-case basis like this, or by creating a theme object that you can make the default in your environment.
Idioms to Make Life Easier
"contractions" and "slang"
typical usage
p_typ <- ggplot(bb_long_al, aes(x=year, y=value, color=stat)) +
geom_line() +
geom_point(size=2)+
ylab("Total") + xlab("Year") +
scale_color_manual("Stat", values=c('darkgreen','darkblue','darkred')) +
ggtitle("American League") + theme_bw()
quick note on stats
simplest faceted plot
p_typ <- ggplot(bb_long, aes(x=year, y=value, color=stat)) +
geom_line() +
facet_wrap(~lg)
stats, annotations and alpha
p_annot <- ggplot(bb_long, aes(x=year, y=value, color=stat)) +
geom_line(alpha=.2, size=2) +
stat_smooth() +
facet_wrap(~lg) +
annotate("text", x=1981, y=0, label="Strike!",
angle=90, hjust=0)
other fun examples
nicely formatted time series
library(scales)
df <- data.frame(
date = seq(Sys.Date(), len=100, by="1 day")[sample(100, 50)],
price = runif(50)
)
df <- df[order(df$date), ]
p_ts <- ggplot(df, aes(date, price)) + geom_line() +
scale_x_date(breaks=date_breaks("2 weeks"))
you are my density
p_density <- ggplot(baseball, aes(ab, fill=lg)) +
geom_density(alpha=.4)
fizzy bubbly plots
library(Hmisc)
p_fb <- ggplot(baseball, aes(lg, ab)) +
geom_jitter() +
stat_summary(fun.data='median_hilow',geom='crossbar', color='red', conf.int=.9)
