Today we’ll look at GGPlot2, the graphics package associated with the Tidyverse.
You will be introduced into how and why to use visualizations in DS 6001.
Visualizations in EDA
Visualization in creating data products that communicate results, such as scientific publications, infographics, and interactive visualizations.
These things come at the end of the data science pipeline.
Today, I just want to introduce you to the thinking and design logic behind the package, so you can be confident in learning more as you need to.
As with Dplyr, GGPlot2 is a an entirely new system that supplants the older graph functions that are built into R.
And just as with Dplyr, it is founded on a principled analysis of its domain and approaches code design through developing a basic grammar which can then be expressed in R.
In effect, Dplyr is built on a grammar of data by defining a set of verbs that can be used to build phrases that are put together into larger constructs.
These verbs correspond to a process of data transformation.
GGPlot2 is built on a grammar of graphics that defines a set of nouns that correspond to the architecture of a graphic (aka plot).
The phrase “grammar of graphics” actually comes from the book by that name written by statistician and computer scientist Leland Wilkinson in 1999 and later revised:
It’s worth reading if you want to get a solid grounding in visualization, which belongs to the design area of data science.
Wilkinson takes an object-oriented approach to visualization and formalizes two main principles:
The essential grammatical elements to create any visualization are:
According to Wickham, who adopted these principles and applied them to R,
A grammar of graphics is a tool that enables us to concisely describe the components of a graphic. Such a grammar allows us to move beyond named graphics (e.g., the “scatterplot”) and gain insight into the deep structure that underlies statistical graphics (Wickham 2012).
Wickham takes this idea and develops it into this:
Source (see also ScienceCraft).
You can see that everything starts with data.
Then data are mapped on aesthetics within geometries.
Geometries are geometric things like points, lines, and bars.
Aesthetics are visual things like position, size, color and shape.
You can see how the latter are properties of the former.
Also note that aesthetics make use of visual channels to signify
Size can means greater than, which is good for numeric scale but not categories
Color can signify things like value, e.g. via red : dangerous : : green : safe.
These are the primary layers. The other layers apply downstream modifications that add more information and style to the graph.
Everything starts with ggplot() which is part of the Tidyverse.
library(tidyverse)── Attaching packages ─────────────────────────────────────── tidyverse 1.3.2 ──
✔ ggplot2 3.3.5 ✔ purrr 1.0.1
✔ tibble 3.2.1 ✔ dplyr 1.1.1
✔ tidyr 1.2.1 ✔ stringr 1.5.0
✔ readr 2.1.3 ✔ forcats 1.0.0
── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
✖ dplyr::filter() masks stats::filter()
✖ dplyr::lag() masks stats::lag()Here is a basic graph – a scatterplot comparing two features in the iris dataset. I’ve broken out the functions and arguments so you can see how the grammar is implemented:
iris %>%
ggplot(
mapping = aes(
x = Sepal.Length,
y = Sepal.Width)
) +
geom_point(size=3, aes(color=Species))
ggplot() starts by creating a coordinate system that you can add layers to.
These layers are created by geometry functions.
geom_point creates a point-based visualization.geom_ functions, and they can be layered on top of each other:
geom_point()geom_bar()geom_histogram()geom_boxplot()Here we have a plot with two layers. The second layer is created by a stat function, which is similar to geom, but applies a statistical transformation to the data.
iris %>%
ggplot(aes(
x = Sepal.Length,
y = Sepal.Width)) +
geom_point(size=3, aes(color=Species)) +
stat_smooth(method = lm)`geom_smooth()` using formula 'y ~ x'
The core the process is that each layer maps data onto what are called aesthetics (aes).
Aesthetics are visual objects and properties that can used to represent numeric and categorical values
In addition to these elements, ggplot also provides faceting, which is the visual equivalent of grouping by. Just as with group by, a data feature is used to divide the visualization into groups, each taking the same form but showing a different subset of data.
iris %>%
ggplot(aes(
x = Sepal.Length,
y = Sepal.Width)) +
geom_point(size=3, aes(color=Species)) +
stat_smooth(method = lm) +
facet_wrap(facets = vars(Species))`geom_smooth()` using formula 'y ~ x'
By the way, this is an example of Simpson’s Paradox. The overall trend is downward, but each group trend upward.
iris %>%
ggplot(aes(
x = Sepal.Length,
y = Sepal.Width)) +
geom_point(size=3, aes(color=Species)) +
stat_smooth(method = lm) +
stat_smooth(method = lm, se=FALSE, aes(color=Species))`geom_smooth()` using formula 'y ~ x'
`geom_smooth()` using formula 'y ~ x'
Anyway, the general structure of a ggplot statement is the following:
ggplot(data = <DATA>) +
<GEOM_FUNCTION>(
mapping = aes(<MAPPINGS>),
stat = <STAT>,
position = <POSITION>
) +
<COORDINATE_FUNCTION> +
<FACET_FUNCTION>+ operatorYou will notice the use of the + operator to connect ggplot functions together to produce a final product. Theses are not quite the same as pipes %>%.
The difference is that pipes feed data from one function to another, whereas the + operation combines elements to produce an increasingly developed visualization.
Another thing to keep in mind: the + always goes at the end of a line, not at the beginning.
Let look at how to build out graphics using the built-in diamonds data.
diamonds# A tibble: 53,940 × 10
carat cut color clarity depth table price x y z
<dbl> <ord> <ord> <ord> <dbl> <dbl> <int> <dbl> <dbl> <dbl>
1 0.23 Ideal E SI2 61.5 55 326 3.95 3.98 2.43
2 0.21 Premium E SI1 59.8 61 326 3.89 3.84 2.31
3 0.23 Good E VS1 56.9 65 327 4.05 4.07 2.31
4 0.29 Premium I VS2 62.4 58 334 4.2 4.23 2.63
5 0.31 Good J SI2 63.3 58 335 4.34 4.35 2.75
6 0.24 Very Good J VVS2 62.8 57 336 3.94 3.96 2.48
7 0.24 Very Good I VVS1 62.3 57 336 3.95 3.98 2.47
8 0.26 Very Good H SI1 61.9 55 337 4.07 4.11 2.53
9 0.22 Fair E VS2 65.1 61 337 3.87 3.78 2.49
10 0.23 Very Good H VS1 59.4 61 338 4 4.05 2.39
# ℹ 53,930 more rowsdiamonds %>%
ggplot(aes(x=carat, y=price)) +
geom_point()
Here’s another way to construct a graphic. By assigning it a variable, we can keep adding to it and showing out work.
gg <- ggplot(diamonds, aes(x=carat, y=price))
gg + geom_point()
stroke controls the thickness of point boundary
gg + geom_point(
size=1,
shape=1,
color="steelblue",
stroke=2) 
Let’s map the variables carat, cut and color to various aesthetics in our geometry function:
gg + geom_point(aes(
size=carat,
shape=cut,
color=color,
stroke=carat))Warning: Using shapes for an ordinal variable is not advised
labs()gg1 <- gg + geom_point(aes(color=color))
gg2 <- gg1 + labs(title="Diamonds", x="Carat", y="Price")
gg2
theme()gg2 + theme(text=element_text(color="blue")) # all text turns blue.
plot.title: Controls plot title.axis.title.x: Controls X axis titleaxis.title.y: Controls Y axis titleaxis.text.x: Controls X axis textaxis.text.y: Controls y axis textgg3 <- gg2 +
theme(plot.title=element_text(size=25),
axis.title.x=element_text(size=20),
axis.title.y=element_text(size=20),
axis.text.x=element_text(size=15),
axis.text.y=element_text(size=15)
)
gg3
gg3 +
labs(title = "Plot Title\nSecond Line of Plot Title") +
theme(plot.title = element_text(
face="bold",
color="steelblue",
lineheight=1.2)
)
gg3 + scale_colour_manual(
name='Legend',
values=c('D'='grey',
'E'='red',
'F'='blue',
'G'='yellow',
'H'='black',
'I'='green',
'J'='firebrick'))
Method 1: Zoom in
gg3 + coord_cartesian(xlim=c(0,3), ylim=c(0, 5000)) + geom_smooth() # zoom in`geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'
Method 2: Deletes the points outside limits
gg3 +
xlim(c(0,3)) +
ylim(c(0, 5000)) +
geom_smooth() # deletes the points `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'Warning: Removed 14714 rows containing non-finite values (stat_smooth).Warning: Removed 14714 rows containing missing values (geom_point).
Method 3: Deletes the points outside limits
gg3 + scale_x_continuous(limits=c(0,3)) +
scale_y_continuous(limits=c(0, 5000)) +
geom_smooth() # deletes the points outside limits`geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'Warning: Removed 14714 rows containing non-finite values (stat_smooth).Warning: Removed 14714 rows containing missing values (geom_point).
Notice the change in smoothing line because of deleted points. This could sometimes be misleading in your analysis.
gg3 + scale_x_continuous(
labels=c("zero", "one", "two", "three", "four", "five")) 
scale_y_continuous(breaks=seq(0, 20000, 4000)) # if Y is continuous, if X is a factor<ScaleContinuousPosition>
Range:
Limits: 0 -- 1Use scale_x_discrete instead, if X variable is a factor.
Rotate axis text
gg3 + theme(axis.text.x=element_text(angle=45), axis.text.y=element_text(angle=45))
gg3 + coord_flip() # flips X and Y axis.
gg3 + theme(panel.background = element_rect(fill = 'springgreen'),
panel.grid.major = element_line(colour = "firebrick", size=3),
panel.grid.minor = element_line(colour = "blue", size=1))
gg3 + theme(plot.background=element_rect(fill="yellowgreen"), plot.margin = unit(c(2, 4, 1, 3), "cm")) # top, right, bottom, left
gg3 + theme(legend.position="none") # hides the legend
gg3 + scale_color_discrete(name="") # Remove legend title (method1)
# Remove legend title (method)
p1 <- gg3 + theme(legend.title=element_blank())
# Change legend title
p2 <- gg3 + scale_color_discrete(name="Diamonds") # install.packages("gridExtra")library(gridExtra)
Attaching package: 'gridExtra'The following object is masked from 'package:dplyr':
combinegrid.arrange(p1, p2, ncol=2) # arrange
gg3 + scale_colour_manual(name='Legend', values=c('D'='grey', 'E'='red', 'F'='blue', 'G'='yellow', 'H'='black', 'I'='green', 'J'='firebrick'))
Outside plot
p1 <- gg3 + theme(legend.position="top") # top / bottom / left / rightInside plot
p2 <- gg3 + theme(legend.justification=c(1,0), legend.position=c(1,0)) # legend justification is the anchor point on the legend, considering the bottom left of legend as (0,0)
gridExtra::grid.arrange(p1, p2, ncol=2)
#df$newLegendColumn <- factor(df$legendcolumn, levels=c(new_order_of_legend_items), ordered = TRUE) Create a new factor variable used in the legend, ordered as you need. Then use this variable instead in the plot.
legend.title - Change legend titlelegend.text - Change legend textlegend.key - Change legend boxguides - Change legend symbolsgg3 + theme(legend.title = element_text(size=20, color = "firebrick"), legend.text = element_text(size=15), legend.key=element_rect(fill='steelblue')) + guides(colour = guide_legend(override.aes = list(size=2, shape=4, stroke=2))) # legend title color and size, box color, symbol color, size and shape.
# Not Run: gg + geom_text(aes(xcol, ycol, label=round(labelCol), size=3))
# general format
gg + geom_text(aes(label=color, color=color), size=4)
##Annotation
library(grid)
my_grob = grobTree(textGrob("My Custom Text", x=0.8, y=0.2,
gp=gpar(col="firebrick", fontsize=25, fontface="bold")))
gg3 + annotation_custom(my_grob)
p1 <- gg1 + facet_grid(color ~ cut) # arrange in a grid. More space for plots. Free X and Y axis scalesBy setting scales=‘free’, the scales of both X and Y axis is freed. Use scales=‘free_x’ to free only X-axis and scales=‘free_y’ to free only Y-axis.
p2 <- gg1 + facet_wrap(color ~ cut, scales="free") # free the x and yaxis scales. library(gridExtra)
grid.arrange(p1, p2, ncol=2)
gg3 + geom_smooth(aes(color=color)) # method could be - 'lm', 'loess', 'gam'`geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'
p1 <- gg3 + geom_hline(yintercept=5000, size=2, linetype="dotted", color="blue") # linetypes: solid, dashed, dotted, dotdash, longdash and twodash
p2 <- gg3 + geom_vline(xintercept=4, size=2, color="firebrick")
p3 <- gg3 + geom_segment(aes(x=4, y=5000, xend=4, yend=10000, size=2, lineend="round")) Warning: Ignoring unknown aesthetics: lineendp4 <- gg3 + geom_segment(aes(x=carat, y=price,
xend=carat, yend=price-500, color=color), size=2) + coord_cartesian(xlim=c(3, 5)) # x, y: start points. xend, yend: endpoints
gridExtra::grid.arrange(p1,p2,p3,p4, ncol=2)
# Frequency bar chart: Specify only X axis.
gg <- ggplot(mtcars, aes(x=cyl))
gg + geom_bar() # frequency table
gg <- ggplot(mtcars, aes(x=cyl))
p1 <- gg + geom_bar(position="dodge", aes(fill=factor(vs))) # side-by-side
p2 <- gg + geom_bar(aes(fill=factor(vs))) # stacked
gridExtra::grid.arrange(p1, p2, ncol=2)
# Absolute bar chart: Specify both X adn Y axis. Set stat="identity"
df <- aggregate(mtcars$mpg, by=list(mtcars$cyl), FUN=mean) # mean of mpg for every 'cyl'
names(df) <- c("cyl", "mpg")
head(df) cyl mpg
1 4 26.66364
2 6 19.74286
3 8 15.10000#> cyl mpg
#> 1 4 26.66
#> 2 6 19.74
#> 3 8 15.10
gg_bar <- ggplot(df, aes(x=cyl, y=mpg)) + geom_bar(stat = "identity") # Y axis is explicit. 'stat=identity'
print(gg_bar)
gg_bar <- ggplot(df, aes(x=cyl, y=mpg)) + geom_bar(stat = "identity", aes(fill=cyl))
print(gg_bar)
df$cyl <- as.factor(df$cyl)
gg_bar <- ggplot(df, aes(x=cyl, y=mpg)) + geom_bar(stat = "identity", aes(fill=cyl), width = 0.25)
gg_bar + scale_fill_manual(values=c("4"="steelblue", "6"="firebrick", "8"="darkgreen"))
library(RColorBrewer)Warning: package 'RColorBrewer' was built under R version 4.0.5display.brewer.all(n=20, exact.n=FALSE) # display available color palettes
ggplot(mtcars, aes(x=cyl, y=carb, fill=factor(cyl))) + geom_bar(stat="identity") + scale_fill_brewer(palette="Reds") # "Reds" is palette name
# Method 1:
gg <- ggplot(economics, aes(x=date)) # setup
gg + geom_line(aes(y=psavert), size=2, color="firebrick") + geom_line(aes(y=uempmed), size=1, color="steelblue", linetype="twodash") # No legend
# available linetypes: solid, dashed, dotted, dotdash, longdash and twodash# Method 2:
#install.packages("reshape2")
library(reshape2)
Attaching package: 'reshape2'The following object is masked from 'package:tidyr':
smithsdf_melt <- melt(economics[, c("date", "psavert", "uempmed")], id="date") # melt by date.
gg <- ggplot(df_melt, aes(x=date)) # setup
gg + geom_line(aes(y=value, color=variable), size=1) + scale_color_discrete(name="Legend") # gets legend.
# One step method.
# install.packages("ggfortify")
library(ggfortify)Warning: package 'ggfortify' was built under R version 4.0.5autoplot(AirPassengers, size=2) + labs(title="AirPassengers")
Filled time series can be plotted using geom_ribbon(). It takes two compulsory arguments ymin and ymax.
# Prepare the dataframe
st_year <- start(AirPassengers)[1]
st_month <- "01"
st_date <- as.Date(paste(st_year, st_month, "01", sep="-"))
dates <- seq.Date(st_date, length=length(AirPassengers), by="month")
df <- data.frame(dates, AirPassengers, AirPassengers/2)
head(df) dates AirPassengers AirPassengers.2
1 1949-01-01 112 56.0
2 1949-02-01 118 59.0
3 1949-03-01 132 66.0
4 1949-04-01 129 64.5
5 1949-05-01 121 60.5
6 1949-06-01 135 67.5#> dates AirPassengers AirPassengers.2
#> 1 1949-01-01 112 56.0
#> 2 1949-02-01 118 59.0
#> 3 1949-03-01 132 66.0
#> 4 1949-04-01 129 64.5
#> 5 1949-05-01 121 60.5
#> 6 1949-06-01 135 67.5
# Plot ribbon with ymin=0
gg <- ggplot(df, aes(x=dates)) + labs(title="AirPassengers") + theme(plot.title=element_text(size=30), axis.title.x=element_text(size=20), axis.text.x=element_text(size=15))
gg + geom_ribbon(aes(ymin=0, ymax=AirPassengers)) + geom_ribbon(aes(ymin=0, ymax=AirPassengers.2), fill="green")
gg + geom_ribbon(aes(ymin=AirPassengers-20, ymax=AirPassengers+20)) + geom_ribbon(aes(ymin=AirPassengers.2-20, ymax=AirPassengers.2+20), fill="green")Don't know how to automatically pick scale for object of type ts. Defaulting to continuous.
geom_area is similar to geom_ribbon, except that the ymin is set to 0. If you want to make overlapping area plot, use the alpha aesthetic to make the top layer translucent.
# Method1: Non-Overlapping Area
df <- reshape2::melt(economics[, c("date", "psavert", "uempmed")], id="date")
head(df, 3) date variable value
1 1967-07-01 psavert 12.6
2 1967-08-01 psavert 12.6
3 1967-09-01 psavert 11.9#> date variable value
#> 1 1967-07-01 psavert 12.5
#> 2 1967-08-01 psavert 12.5
#> 3 1967-09-01 psavert 11.7
p1 <- ggplot(df, aes(x=date)) + geom_area(aes(y=value, fill=variable)) + labs(title="Non-Overlapping - psavert and uempmed")
# Method2: Overlapping Area
p2 <- ggplot(economics, aes(x=date)) + geom_area(aes(y=psavert), fill="yellowgreen", color="yellowgreen") + geom_area(aes(y=uempmed), fill="dodgerblue", alpha=0.7, linetype="dotted") + labs(title="Overlapping - psavert and uempmed")
gridExtra::grid.arrange(p1, p2, ncol=2)
The oulier points are controlled by the following aesthetics:
If the notch is turned on (by setting it TRUE), the below boxplot is produced. Else, you would get the standard rectangular boxplots.
p1 <- ggplot(mtcars, aes(factor(cyl), mpg)) + geom_boxplot(aes(fill = factor(cyl)), width=0.5, outlier.colour = "dodgerblue", outlier.size = 4, outlier.shape = 16, outlier.stroke = 2, notch=T) + labs(title="Box plot") # boxplot
p2 <- ggplot(mtcars, aes(factor(cyl), mpg)) + geom_violin(aes(fill = factor(cyl)), width=0.5, trim=F) + labs(title="Violin plot (untrimmed)") # violin plot
gridExtra::grid.arrange(p1, p2, ncol=2)notch went outside hinges. Try setting notch=FALSE.
notch went outside hinges. Try setting notch=FALSE.
ggplot(mtcars, aes(mpg)) + geom_density(aes(fill = factor(cyl)),
size=2) + labs(title="Density plot") 
corr <- round(cor(mtcars), 2)
df <- reshape2::melt(corr)
gg <- ggplot(df, aes(x=Var1, y=Var2, fill=value, label=value)) + geom_tile() + theme_bw() + geom_text(aes(label=value, size=value), color="white") + labs(title="mtcars - Correlation plot") + theme(text=element_text(size=20), legend.position="none")
library(RColorBrewer)
p2 <- gg + scale_fill_distiller(palette="Reds")
p3 <- gg + scale_fill_gradient2()
gridExtra::grid.arrange(gg, p2, p3, ncol=3)