Course 4 - Exploratory Data Analysis - Week 2 - Notes

Lattice Plotting System

Useful for plotting high-dimensional data. Doesn’t have atwo phase plotting - everything is in one call.

Functions

• xyplot() for scatterplots.
• bwplot() for boxplots.
• histogram() for histograms.
• stripplot() like a boxplot but with actual points.
• dotplot() to plot does on ‘violin strings’.
• splom() for a scatterplot matrix - like pairs() in the base system.
• levelplot(), contourplot() for plotting ‘image’ data.

The functions generally takes a formula with some conditioning variables. e.g.

# We have <y-axis ~ x-axis> on the left.
# f and g are conditioning variables - they are optional.
xyplot(y ~ x | f * g, data)

If no data frame is supplied, it will look in the parent environment frame for those variables.

Simple Plot

library(lattice)
library(datasets)
xyplot(Ozone ~ Wind, airquality)

Can separate into different plots based on a categorical variable:

airquality <- airquality %>% mutate(Month = as.factor(Month))
xyplot(Ozone ~ Wind | Month, airquality, layout = c(5,1))

Behaviour

The lattice functions behave differently from base in one important way:

• Base plots data directly to a graphics device.
• Lattice returns an object of class trellis.
• The print methods for lattice functions actually plot the data on the graphics device.
• On the command line, the trellis objects are auto-printed.

Panel Functions

The lattice functions have a panel function which controls what happens inside each panel of the plot. It comes with default functions, but these can be overridden.

The panel functions receive x/y coordinates of the data points in their panel along with optional arguments.

Each panel represents a subset of the data based on the conditioning variable passed to the function.

We can plot with two panels:

set.seed(1)
raw <- tibble(
    x = rnorm(100),
    f = rep(0:1, each = 50),
    y = x + f - f * x + rnorm(100, sd = 0.5)
) %>%
    mutate(f = factor(f, labels = c('G1', 'G2')))

xyplot(y ~ x | f, raw, layout = c(2,1)) 

Now we provide a custom panel function. It first calls the default panel function, then adds a horizontal line at the median.

xyplot(y ~ x | f, data = raw, panel = function(x,y, ...) {
    panel.xyplot(x,y,...)
    panel.abline(h = median(y), lty = 2)
})

We could also add a regression line. The col arugment specifies the colour.

xyplot(y ~ x | f, data = raw, panel = function(x,y, ...) {
    panel.xyplot(x,y,...)
    panel.lmline(x, y, col = 2)
})

Summary

• Single function call.
• Aspects like margins and spacing are automatically handled.
• Ideal for creating plots where the same plot is viewed under many conditions.
• Panel functions are used to customise what is plotted in each panel.

ggplot2

An implementation of Grammar of Graphics by Leland Wilkinson. Written by Hadley Wickham.

… the grammar tells us that a statistical graphic is a mapping from the data to aesthetic attributes (colour, shape, size) of geometric objects (points, lines). The plot may also contain statistical transformations of the data, and is drawn on a specific coordinate system.

Basics

The qplot() function works much like plot() in base. Looks for data in a data frame, or parent environment. Plots are made up of aesthetics and geoms.

ggplot() is the core function and very flexible for doing things qplot() cannot do.

The components are:

• A data frame.
• Aesthetic mappings: how data are mapped to colour, size.
• Geoms: geometric objects like points, lines, shapes.
• Facets: for conditional plots.
• Stats: statistical transformations like binning, quantiles, smoothing.
• Scales: what scale an aesthetic map uses.
• Coordinate system.

Outliers

If you use + ylim(), the outlier will be missing. If you use + coord_cartesian(ylim = c(min, max)), the outlier will be included.