Last week: Text manipulation

Part I

Plot basics

Plotting in R

Base R has a set of powerful plotting tools. An overview:

The ggplot2 package also provides very nice (and very different) plotting tools; we won’t cover it in this course (it tends to be the focus in Statistical Graphics, 36-315)

Scatter plot

To make a scatter plot of one variable versus another, use plot()

n = 50
set.seed(0)
x = sort(runif(n, min=-2, max=2))
y = x^3 + rnorm(n)
plot(x, y)

Plot type

The type argument controls the plot type. Default is p for points; set it to l for lines

plot(x, y, type="p")

plot(x, y, type="l")

Try also b or o, for both points and lines

Labels

The main argument controls the title; xlab and ylab are the x and y labels

plot(x, y, main="A noisy cubic") # Note the default x and y labels

plot(x, y, main="A noisy cubic", xlab="My x variable", ylab="My y variable")

Point type

Use the pch argument to control point type

plot(x, y, pch=21) # Empty circles, default

plot(x, y, pch=19) # Filled circles

Try also 20 for small filled circles, or "." for single pixels

Line type

Use the lty argument to control the line type, and lwd to control the line width

plot(x, y, type="l", lty=1, lwd=1) # Solid line, default width

plot(x, y, type="l", lty=2, lwd=3) # Dashed line, 3 times as thick

Color

Use the col argument to control the color. Can be:

The function colors() returns a string vector of the available colors

plot(x, y, pch=19, col=1) # Black, default

plot(x, y, pch=19, col=2) # Red

Multiple plots

To set up a plotting grid of arbitrary dimension, use the par() function, with the argument mfrow. Note: in general this will affect all following plots! (Except in separate R Markdown code chunks …)

par(mfrow=c(2,2)) # Grid elements are filled by row
plot(x, y, main="Red cubic", pch=20, col="red")
plot(x, y, main="Blue cubic", pch=20, col="blue")
plot(rev(x), y, main="Flipped green", pch=20, col="green")
plot(rev(x), y, main="Flipped purple", pch=20, col="purple")

Margin

Default margins in R are large (and ugly); to change them, use the par() function, with the argument mar. Note: in general this will affect all following plots! (Except in separate R Markdown code chunks …)

par(mfrow=c(2,2), mar=c(4,4,2,0.5))
plot(x, y, main="Red cubic", pch=20, col="red")
plot(x, y, main="Blue cubic", pch=20, col="blue")
plot(rev(x), y, main="Flipped green", pch=20, col="green")
plot(rev(x), y, main="Flipped purple", pch=20, col="purple")

# Evidence that par() does not carry over to separate R Markdown code chunks 
plot(x, y)

Saving plots

Use the pdf() function to save a pdf file of your plot, in your R working directory. Use getwd() to get the working directory, and setwd() to set it

getwd() # This is where the pdf will be saved
## [1] "/Users/ryantibshirani/Dropbox/teaching/f21-350/lectures/plotting"
pdf(file="noisy_cubics.pdf", height=7, width=7) # Height, width are in inches
par(mfrow=c(2,2), mar=c(4,4,2,0.5))
plot(x, y, main="Red cubic", pch=20, col="red")
plot(x, y, main="Blue cubic", pch=20, col="blue")
plot(rev(x), y, main="Flipped green", pch=20, col="green")
plot(rev(x), y, main="Flipped purple", pch=20, col="purple")
graphics.off()

Also, use the jpg() and png() functions to save jpg and png files

Adding to plots

The main tools for this are:

You’ll get practice with this on lab. Pay attention to layers—they work just like they would if you were painting a picture by hand

Part II

Histograms and heatmaps

Plotting a histogram

To plot a histogram of a numeric vector, use hist()

king.lines =
  readLines("http://www.stat.cmu.edu/~ryantibs/statcomp/data/king.txt")
king.words = strsplit(paste(king.lines, collapse=" "),
                      split="[[:space:]]|[[:punct:]]")[[1]]
king.words = tolower(king.words[king.words != ""])
king.wlens = nchar(king.words)
hist(king.wlens)

Histogram options

Several options are available as arguments to hist(), such as col, freq, breaks, xlab, ylab, main

hist(king.wlens, col="pink", freq=TRUE) # Frequency scale, default

hist(king.wlens, col="pink", freq=FALSE, # Probability scale, and more options
     breaks=0:20, xlab="Word length", main="King word lengths")

Adding a histogram to an existing plot

To add a histogram to an existing plot (say, another histogram), use hist() with add=TRUE

hist(king.wlens, col="pink", freq=FALSE, breaks=0:20, 
     xlab="Word length", main="King word lengths")
hist(king.wlens + 5, col=rgb(0,0.5,0.5,0.5), # Note: using a transparent color
     freq=FALSE, breaks=0:20, add=TRUE)

Adding a density curve to a histogram

To estimate a density from a numeric vector, use density(). This returns a list; it has components x and y, so we can actually call lines() directly on the returned object

density.est = density(king.wlens, adjust=1.5) # 1.5 times the default bandwidth
class(density.est)
## [1] "density"
names(density.est)
## [1] "x"         "y"         "bw"        "n"         "call"      "data.name" "has.na"
hist(king.wlens, col="pink", freq=FALSE, breaks=0:20, 
     xlab="Word length", main="King word lengths")
lines(density.est, lwd=3)

Plotting a heatmap

To plot a heatmap of a numeric matrix, use image()

(mat = 1:5 %o% 6:10) # %o% gives for outer product
##      [,1] [,2] [,3] [,4] [,5]
## [1,]    6    7    8    9   10
## [2,]   12   14   16   18   20
## [3,]   18   21   24   27   30
## [4,]   24   28   32   36   40
## [5,]   30   35   40   45   50
image(mat) # Red means low, white means high

Orientation of image()

The orientation of image() is to plot the heatmap according to the following order, in terms of the matrix elements:

\[\begin{array}{cccc} (1,\text{ncol}) & (2, \text{ncol}) & \ldots & (\text{nrow},\text{ncol}) \\ \vdots & & & \\ (1,2) & (2,2) & \ldots & (\text{nrow},2) \\ (1,1) & (2,1) & \ldots & (\text{nrow},1) \end{array}\]

This is a 90 degrees counterclockwise rotation of the “usual” printed order for a matrix:

\[\begin{array}{cccc} (1,1) & (1,2) & \ldots & (1,\text{ncol}) \\ (2,1) & (2,2) & \ldots & (2,\text{ncol}) \\ \vdots & & & \\ (\text{nrow},1) & (\text{nrow},2) & \ldots & (\text{nrow},\text{ncol}) \end{array}\]

Therefore, if you want the displayed heatmap to follow the usual order, you must rotate the matrix 90 degrees clockwise before passing it in to image(). (Equivalently: reverse the row order, then take the transpose.) Convenient way of doing so:

clockwise90 = function(a) { t(a[nrow(a):1,]) } # Handy rotate function
image(clockwise90(mat))

Color scale

The default is to use a red-to-white color scale in image(). But the col argument can take any vector of colors. Built-in functions gray.colors(), rainbow(), heat.colors(), topo.colors(), terrain.colors(), cm.colors() all return continguous color vectors of given length

phi = dnorm(seq(-2,2,length=50))
normal.mat = phi %o% phi
image(normal.mat) # Default is col = hcl.colors(12, "YlOrRd", rev=TRUE)

image(normal.mat, col=heat.colors(12)) # This was the old default!

image(normal.mat, col=terrain.colors(20)) # Terrain colors

image(normal.mat, col=topo.colors(20)) # Topological colors

Drawing contour lines

To draw contour lines from a numeric matrix, use contour(); to add contours to an existing plot (like, a heatmap), use contour() with add=TRUE

contour(normal.mat)

image(normal.mat, col=terrain.colors(20))
contour(normal.mat, add=TRUE)

Part III

Curves, surfaces, and colors

Drawing a curve

To draw a curve of a function, use curve()

curve(x^3) # Default is to plot between 0 and 1. Note: x here is a symbol

curve(x^3, from=-3, to=3, lwd=3, col="red") # More plotting options

Adding a curve to an existing plot

To add a curve to an existing plot, use curve() with add=TRUE

n = 50
set.seed(0)
x = sort(runif(n, min=-2, max=2))
y = x^3 + rnorm(n)

plot(x, y)
curve(x^3, lwd=3, col="red", add=TRUE)

# Note: the x argument here and the x vector we defined above are different!
# Reminder: x here is a symbol

Adding a rug to an existing plot

To add a rug to an existing plot (just tick marks, for where the x points occur), use rug()

plot(x, y)
curve(x^3, lwd=3, col="red", add=TRUE) 
rug(x)

Drawing a surface

To draw a surface, use surface(), available at http://www.stat.cmu.edu/~ryantibs/statcomp/lectures/surface.R. (This is a function written by Professor Tibs, relying on the built-in persp() function)

source("http://www.stat.cmu.edu/~ryantibs/statcomp/lectures/surface.R")
surface(x^3 + y^3, from.x=-3, to.x=3, from.y=-3, to.y=3)

surface(x^3 + y^3, from.x=-3, to.x=3, from.y=-3, to.y=3,
        theta=25, phi=15, col=terrain.colors(30), 
        ticktype="detailed", mar=c(2,2,2,2))

Adding points to a surface

To add points to a surface, save the output of surface(). Then use trans3d(), to transform (x,y,z) coordinates to (x,y) coordinates that you can pass to points()

persp.mat = surface(x^3 + y^3, from.x=-3, to.x=3, from.y=-3, to.y=3,
                    theta=25, phi=15, col=rgb(0,0,1,alpha=0.2),
                    ticktype="detailed", mar=c(2,2,2,2))

n = 500
x = runif(n, -3, 3)
y = runif(n, -3, 3)
z = x^3 + y^3 + 5*rnorm(n)

xy.list = trans3d(x, y, z, persp.mat)
points(xy.list, pch=20)

Color palettes

Color palettes are functions for creating vectors of contiguous colors, just like gray.colors(), rainbow(), heat.colors(), terrain.colors(), topo.colors(), cm.colors(). Given a number n, each of these functions just returns a vector of colors (names, stored as strings) of length n

n = 50
plot(0, 0, type="n", xlim=c(1,n), ylim=c(1,6))
points(1:n, rep(6,n), col=gray.colors(n), pch=19)
points(1:n, rep(5,n), col=rainbow(n), pch=19)
points(1:n, rep(4,n), col=heat.colors(n), pch=19)
points(1:n, rep(3,n), col=terrain.colors(n), pch=19)
points(1:n, rep(2,n), col=topo.colors(n), pch=19)
points(1:n, rep(1,n), col=cm.colors(n), pch=19)

Creating a custom color palette

To create a custom palette, that interpolates between a set of base colors, colorRampPalette()

cust.colors = colorRampPalette(c("red","purple","darkgreen"))
class(cust.colors)
## [1] "function"
plot(1:n, rep(1,n), col=cust.colors(n), pch=19)

Coloring points by value

Coloring points according to the value of some variable can just be done with a bit of indexing, and the tools you already know about colors

# Function to retrieve a color according to a value
# - val: the value in question
# - lim: a vector of length 2, lower and upper limits for possible values
# - col.vec: the color vector to choose from
get.col.from.val = function(val, lim, col.vec) {
  col.vec[(val-lim[1])/(lim[2]-lim[1]) * (length(col.vec)-1) + 1]
}

# Let's color points according to y value
col.vec = heat.colors(30)
lim = c(-1, 1)
theta = seq(0, 6*pi, length=200)

plot(theta, sin(theta), type="o", pch=19, 
     col=get.col.from.val(sin(theta), lim, col.vec))

# Another example, now in 3d
persp.mat = surface(x^3 + y^3, from.x=-3, to.x=3, from.y=-3, to.y=3,
                    theta=25, phi=15, col=rgb(1,1,1,alpha=0.2),
                    ticktype="detailed", mar=c(2,2,2,2))

# Let's color points according to z value
col.vec = terrain.colors(30)
lim = c(min(z), max(z))

xy.list = trans3d(x, y, z, persp.mat)
points(xy.list, pch=20, col=get.col.from.val(z, lim, col.vec))

Summary