2016.05.26

STS 98 - Lecture 2016.05.26

See notes.R to follow along in RStudio.

Also see the R input from the lecture.

Announcements

Please vote on topics for next Tuesday! (Piazza poll will go up later today).

Lattice Customization

Last time we made a lattice plot of vehicle post frequency by drivetrain.

What if we want to change the line colors and styles?

What if we want to add other lines?

Here's a strategy for making customized lattice plots:

# 0. Compute the data frame or matrix you want to plot.
v = readRDS("cl_vehicles.rds")
drive_freq = table(as.Date(v$date_posted), v$drive)
drive_freq = as.data.frame(drive_freq)
names(drive_freq) = c("date", "drive", "freq")

# 1. Change aesthetic (color, size, etc) settings for the plot and legend.
settings = trellis.par.get()
settings$superpose.line$lty = c("solid", "dashed", "dotted")

library("RColorBrewer")
display.brewer.all()
settings$superpose.line$col =  brewer.pal(3, "Set1")

# trellis.par.set(settings)
key = list(points = FALSE, lines = TRUE)

# 2. (optional) Define a panel function.
panel = function(x, y, ...) {
  # Draw stuff!
  panel.abline(v = as.Date("2016-04-30"), lwd = "3")
  panel.xyplot(x, y, ...)
  panel.points(as.Date("2016-04-28"), 200)
}

# 3. Draw the plot!
library("lattice")
xyplot(freq ~ as.Date(date), drive_freq, groups = drive,
       type = 'l', auto.key = key, par.settings = settings,
       panel = panel)
#abline(v = as.Date("2016-04-30"))

To change styles: change default plot settings in one place, so that the aesthetics are consistent.

show.settings()
settings = trellis.par.get()

To add lines/points: write your own panel function.

Each box or panel in a lattice plot is drawn by a panel function. Lattice has lots of built in panel functions that you can use as building blocks to write your own.

The syntax is:

panel = function(x, y, ...) {
  # Draw stuff!
}

Built-in panel functions:

Function	What does it draw?
panel.3dscatter
panel.3dwire
panel.abline	straight lines
panel.arrows	arrows
panel.average
panel.axis
panel.barchart	bar plot
panel.brush.splom
panel.bwplot	box plot
panel.cloud
panel.contourplot	contour plot
panel.curve	mathematical functions
panel.densityplot	density plot
panel.dotplot	dot plot
panel.error
panel.fill
panel.grid
panel.histogram	histogram
panel.identify
panel.identify.cloud
panel.identify.qqmath
panel.levelplot
panel.levelplot.raster
panel.linejoin
panel.lines	lines
panel.link.splom
panel.lmline
panel.loess
panel.mathdensity
panel.number
panel.pairs
panel.parallel
panel.points	points
panel.polygon
panel.qq
panel.qqmath
panel.qqmathline
panel.rect
panel.refline
panel.rug
panel.segments
panel.smoothScatter	smooth scatter plot
panel.spline
panel.splom	scatter plot matrix
panel.stripplot
panel.superpose
panel.superpose.2
panel.superpose.plain
panel.text	text
panel.tmd.default
panel.tmd.qqmath
panel.violin	violin plot
panel.wireframe	3-d wireframe plot
panel.xyplot	scatter plot

How to use locator with lattice:

library("grid")
loc = grid.locator()

# Use the coordinates in grid drawing functions.

grid.points(loc$x, loc$y)

Let's plot price against odometer again.

# 0. Compute data.

# 1. Change plot settings.

# 2. Define panel.
panel = function(x, y, ...) {
  panel.smoothScatter(x, y, ...)
}

# 3. Draw plot.
xyplot(price ~ odometer | drive, v, xlim = c(0, 1e6), ylim = c(0, 1e5),
       panel = panel, nbin = 200, colramp = viridis)

library("viridis")
colramp = function(n) {
  brewer.pal(n, "Set1")
}

ggplot2

The 3 main ways to make plots in R are:

1. base R - plot() and friends

2. lattice

3. ggplot2

Each is incompatible with the others. For instance, base R plot commands don't affect with lattice plots.

ggplot2's fundamental idea is that ANY graphic is composed of a few layers:

Layer	Name	Examples
data	data	any data frame
aes	aesthetics	x and y position, color, line style
geom	geometry	points, lines, bars, boxes
stat	statistics	none, bins, sums, means
scale	scales	axes, legends
coord	coordinates	Cartesian, logarithmic, polar
facet	facets	side-by-side panels

These layers form a descriptive "grammar of graphics".

See the documentation at http://docs.ggplot2.org/.

So how can we actually make a plot?

#install.packages("ggplot2")
library("ggplot2")

head(diamonds)
qplot(carat, price, data = diamonds)

The "q" in qplot() stands for "quick".

Change the geometry to change the type of plot. It's possible to use multiple geometries.

qplot(price, data = diamonds, geom = "density")

qplot(carat, price, data = diamonds, geom = "point")

qplot(carat, price, data = diamonds, geom = c("point", "density_2d"))

qplot(carat, price, data = diamonds, geom = "bin2d", bins = 200)

qplot(carat, price, data = diamonds, geom = "hex", main = "Hex Plot")

What geometries are available?

abline, area, bar, bin2d, blank, boxplot, contour, count, crossbar, curve, density, density2d, density_2d, dotplot, errorbar, errorbarh, freqpoly, hex, histogram, hline, jitter, label, line, linerange, map, path, point, pointrange, polygon, qq, quantile, raster, rect, ribbon, rug, segment, smooth, spoke, step, text, tile, violin, vline

It's also easy to create faceted plots.

qplot(price, data = diamonds, binwidth = 100, facets = clarity ~ cut)

qplot(carat, price, data = diamonds, facets = cut ~ .)

qplot(carat, price, data = diamonds, facets = . ~ cut)

qplot(color, price / carat, data = diamonds, geom = "boxplot",
  facets = . ~ clarity)

ggplot2 distinguishes between mapping an aesthetic to a variable and setting an aesthetic to a constant value. The default is mapping.

Special syntax is needed to set an aesthetic.

qplot(carat, price, data = diamonds, geom = "point", color = cut)

qplot(carat, price, data = diamonds, geom = "point", color = I("blue"),
  size = I(10))

qplot(carat, price, data = diamonds, geom = "point", size = price,
  color = cut, shape = cut)

qplot(carat, price, data = diamonds, geom = "point", size = price,
  color = cut, alpha = I(1 / 100), main = "Price vs Carat")

Although qplot() is very powerful, sometimes you want even more control to customize the plot.

gg = ggplot(diamonds, aes(x = carat, y = price, color = cut))
gg + geom_point()
gg + geom_line()

The ggplot() function just sets up the data and aesthetic layers. It's up to you to add geometries!

Very easy to drop in a different data frame:

gg = ggplot(diamonds, aes(x = carat, y = price, color = cut))
gg = gg + geom_point()

gg %+% subset(diamonds, price >= 10000)
gg %+% subset(diamonds, cut == "Fair")
gg %+% subset(diamonds, depth > 65)

To learn more about how to customize ggplots, read

• "ggplot2" by Hadley Wickham

• "The R Graphics Cookbook" by Winston Chang

The ggplot2 package can be used with the ggmap package to create very nice maps.

Reproducibility

Writing clean, organized code is an essential programming skill. Sooner or later, someone else will want or need to run your code to reproduce your results. Try to make it easy for them!

Use descriptive variable names and comment on your code so that you and others can understand it later.

Organize code into "paragraphs" that perform a single step of the larger computation. Put blank lines between paragraphs. Use functions to organize steps that might be used many times.

Avoid using absolute paths ("C:/Users/My Data/stock_data.rds"). Use a variable to store the path so it can be changed easily, and use paths relative to your R script ("stock_data.rds")

What Next?

• STA 32 - learn intro stats
• STA 141ABC - advanced R
• ECS 10 - intro python and programming

Data studies classes!

• Advanced R visualizations
• Data ethnography (archaeology)
• Data ethics