Using ggplot2
July 2015
Refs:
Hadley Wichham - ggplot2, Elegant Graphics for Data Analysis
Winston Chang - R Graphics Cookbook
ggplot2 implements Wilkinson grammar of graphics that describes and conceptually organizes the features that underlie statistical graphics.
The most important concepts are:
The data (duh!)
Aesthetic mappings describing how the data is mapped to aesthetic atributes (eg, shape, color, horizontal and vertical position)
Geometric objects, geoms, that we see on the graphic (eg, points, lines)
Statistical mappings, stats, summarizing the data (eg, binning, counting)
Scaling mappings, scales, the map the data space into aesthetic space (eg, linear or log axis, legends, how to map colors), there should be a scale for each aesthetic
Coordinate systems, coord, a map from the data coordinates to the graph coordinates (eg, cartesian, log-log, polar)
Facets describing how to plit the data into subsets
library(ggplot2)
library(gridExtra) # for presenting plots side by sideFALSE Loading required package: gridset.seed(101)
d <- diamonds[sample(nrow(diamonds), 100), ] # read some data for next examples
head(d)FALSE carat cut color clarity depth table price x y z
FALSE 20077 1.00 Very Good F VVS2 60.0 62 8553 6.43 6.46 3.87
FALSE 2364 0.90 Premium J VS1 62.3 56 3175 6.23 6.19 3.87
FALSE 38279 0.21 Premium D VS2 59.1 62 386 3.89 3.86 2.29
FALSE 35474 0.39 Ideal G VVS2 61.9 56 902 4.68 4.70 2.90
FALSE 13477 1.15 Ideal I VS2 61.7 57 5534 6.72 6.74 4.15
FALSE 16184 1.10 Premium D SI1 60.7 55 6468 6.74 6.71 4.08qplot
qplot is ggplot2’s plot function.
plot1 <- qplot(carat, price, data = d)
plot2 <- qplot(log(carat), log(price), data = d)
plot3 <- qplot(carat, x * y * z, data = d) # x*y*z gives the volume
grid.arrange(plot1, plot2, plot3, ncol=3)
qplot automates some aesthetics like how to assign colors and shapes to data:
plot1 <- qplot(carat, price, data = d, colour = color, size= carat)
plot2 <- qplot(carat, price, data = d, shape = cut)
grid.arrange(plot1, plot2, ncol=2)
Notice that the plots also come with a legend. For each aesthetic attribute there is a scale function mapping the data values to the aesthetic values. Eg, in the left plot, the attribute D was associated with red.
qplot accepts different types of geometric objects, geoms, which will make it produce different types of graphics. The default is geom="point", ie, the scatterplots we’ve seen. Other objects are possible:
plot1 <- qplot(carat, price, data = d, geom=c("point", "smooth")) # default smooth by loess regression
plot2 <- qplot(carat, price, data = d, geom=c("point", "boxplot"))
plot3 <- qplot(carat, price, data = d, geom=c("point", "line"))
grid.arrange(plot1, plot2, plot3, ncol=3)
Geoms for 1D data:
plot1 <- qplot(carat, data = d, geom="histogram") # continuous values
plot2 <- qplot(color, data = d, geom="bar") # discrete values
grid.arrange(plot1, plot2, ncol=2) 
plot1 <- qplot(carat, data = d, geom="freqpoly")
plot2 <- qplot(carat, data = d, geom="density")
grid.arrange(plot1, plot2, ncol=2) 
Smooth geoms
The smooth geom can be used with different regression methods:
plot1 <- qplot(carat, price, data = d, geom=c("point", "smooth"), method="lm")
plot2 <- qplot(carat, price, data = d, geom=c("point", "smooth"), method="lm", formula=y~poly(x,3)) # polynomial regression
grid.arrange(plot1, plot2, ncol=2)
library(splines) # using natural splines
plot3 <- qplot(carat, price, data = d, geom=c("point", "smooth"), method="lm", formula=y~ns(x,5))
library(MASS) # for robust regression
plot4 <- qplot(carat, price, data = d, geom=c("point", "smooth"), method="rlm")
grid.arrange(plot3, plot4, ncol=2)
Jitter and opacity
Sometimes the datapoints are too many and a direct plot is unable to transmit an appropriate perspective of the data. One tool is to jitter the points (add small random noise so that many equal data points are spread around its center) and/or define an amount of opacity, ie, stating how many points there must be at area so that the graphic plots without transparency.
plot1 <- qplot(carat, price, data = diamonds)
plot2 <- qplot(carat, price, data = diamonds, alpha=I(1/50)) # 100 pts for total opacity
grid.arrange(plot1, plot2, ncol=2)