+ - 0:00:00
Notes for current slide
Notes for next slide

Data analysis II

Visualisation, practice

1 / 81

Laurent Bergé

University of Bordeaux, BxSE

09/12/2021

Outline

  1. Intro
  2. A bit of graphical code in R
  3. A functional approach to graphs
  4. More graphical code in R
2 / 81

A tale of two paths...

3 / 81

Pros and cons: Base R

Pros

  • the principle is easy to grasp: we simply overlay successive forms on top of each other

  • low level operations \(=\) everything is possible

4 / 81

Pros and cons: Base R

Pros

  • the principle is easy to grasp: we simply overlay successive forms on top of each other

  • low level operations \(=\) everything is possible

Cons

  • because everything is an overlay on stg already in place, the first plot is critical: you need to plan everything in advance!!! Doing simple stuff can be surprisingly difficult.

  • many commands which are not really intuitive. After 10 years, I still look up ?par regularly.

4 / 81

Messing up the first plot in Base R

You miss most of the show!

5 / 81

Disclaimer

I'm a ggplot2 noob, base R is my home. I have a lot of sympathy for it.

6 / 81

Pros and cons: ggplot2

Pros

  • much more user friendly: stacks all your layers before creating the graph and makes all the computations for you. You don't need to overthink how to set the stage any more.

  • millions of contributed packages

7 / 81

Pros and cons: ggplot2

Pros

  • much more user friendly: stacks all your layers before creating the graph and makes all the computations for you. You don't need to overthink how to set the stage any more.

  • millions of contributed packages

Cons

  • since there is pre-processing, it is sometimes difficult to do exactly what you want (e.g. for a very precise publication graph). To be confirmed, it's only second-hand experience!
7 / 81

Wait! There are other paths!

8 / 81

Ready made solutions

  • there are numerous packages out there which make good graphs with a user-friendly interface (ie with minimal user input)

  • example: ggpubr or fplot for distributions; ggcorrplot for correlations; highcharter for many things, etc.

9 / 81

Pros and cons: Ready made solutions

Pros

  • in a single line of code you get a graph of (usually) very decent quality

  • very good for exploratory graphs

10 / 81

Pros and cons: Ready made solutions

Pros

  • in a single line of code you get a graph of (usually) very decent quality

  • very good for exploratory graphs

Cons

  • the level of customization is limited, this is especially problematic for presentations/publications in which you want a very high level of customization
10 / 81
\(\star\): Or any equivalent image edition software, but inkscape is both free and powerful (yeah!).

Direct edition

  • I strongly encourage you to learn how to work with inkscape\(\star\)

  • it's just crazy how fast you can edit/create images

  • that's indispensable in your skill set, you'll save so much time!

11 / 81

later: add gif edition of the castle image

Pros and cons: Direct edition

Pros

  • you can do exactly what you want, as precisely as you want

  • with practice, you can edit very rapidly

12 / 81

Pros and cons: Direct edition

Pros

  • you can do exactly what you want, as precisely as you want

  • with practice, you can edit very rapidly

Cons

  • the direct edition only comes after the first creation of the graph: hence you have to navigate across software

  • cannot be automated: all the work that you do with one graph, you'd have to do it again for a graph with new data

12 / 81

Short introduction to Base R

13 / 81

Data content

R dispose of multiple functions to display data:

  • plot(): CORE graphical function
  • points(), lines(), abline(), text(): used to plot additional data
  • density(), hist(), boxplot(), etc...
14 / 81

Plot

  • The function plot() is the main graphical function of R (more precisely, it's a method).

  • By default it is a scatterplot between two variables, but it can be used to do much more than that.

  • Some functions preprocess the data, like density(), and modify completely the behavior of plot() when you apply it to the preprocessed data. More on that later.

  • When you apply plot(), it creates a new graphic and the previous one is lost (of course there are exceptions...). To add several pieces of information, you'll need to use other functions.

15 / 81

Main plot arguments

Main plot arguments relating to data:

  • x, y: the data
  • xlim, ylim: the limits of the plotting region
  • col, pch, lty, lwd: color, symbol, line type and line width
  • type: the type of plot
  • log: whether to put the x/y axes to logarithm
16 / 81

Plot: type

17 / 81

Plot: type = "n"

Using type = "n" hides the data, but EVERYTHING else is there. Can be useful when constructing complex graphics: i.e. when setting the stage.

plot(1:5, type = "n")
18 / 81

Plot: limits

19 / 81

Plot: pch

plot(1:20, pch = 1:20)
grid()
20 / 81

Plot: cex

plot(1:5, pch = 16, cex = 1:5, main = "cex: modify point size")
21 / 81

Plot: lty and lwd

22 / 81

Plot: col I

23 / 81

Plot: col II

Lot of color possibilities:

  • Custom colors: rgb(),hsv(), etc
  • "Nice" colors: package RColorBrewer

  • Color interpolation:

    • rainbow(n), heat.colors(n), etc, create vectors of n colors.
    • colorRampPalette(c("white", "blue"))(5): create a vector of 5 colors between the colors white and blue.

  • Nice introduction to R colors in the R-stats UBC course

24 / 81

Exercise: Plot

Generate a 100 periods Brownian motion \(x_{t+1} = x_{t} + \epsilon_{t}\), \(\epsilon_{t}\sim N(0,1)\).

  1. Plot its evolution with both a solid line and filled points (in the same graph).
  2. This time display only the points and use the function rainbow() to set the color of each point.
25 / 81

Adding data points

To add points/lines onto an existing plot:

  • lines()
  • points()

It behaves as the function plot() and contains the same arguments (col, lty, cex, lwd, pch).

26 / 81

Lines & points

plot(1:5, ylim = c(-2, 5))
lines(1:5 - 1)
points(1:5 - 2)
27 / 81

Exercise: Plot & line

In the following graph, the functions plot(), lines() and points() have been called. Can you say to what command refers each graphical information, and in what order they have been called?

28 / 81

Exercise: Plot & lines

Re-generate the previous Brownian motion.

  1. Plot it with both line and dots.

  2. Generate another Brownian motion with \(\epsilon_{t}\sim N(0, 4)\).
    Plot the two motions on a single graph, the second one should be of "firebrick" color, have thick and dashed line and be of triangle symbol.

29 / 81

abline I

The function abline() draws lines. Its arguments are:

  • h: coordinate of horizontal line
  • v: coordinate of vertical line
  • a, b: intercept (a) and slope of a straight line. Shorthan exist: can take the result of an OLS regression (function lm()) instead.
30 / 81

abline II

plot(iris$Sepal.Length, iris$Petal.Width)
abline(lm(Petal.Width ~ Sepal.Length, iris))
abline(h = c(1, 2), v = c(5, 7), col = "gray", lty = 3)
31 / 81

Exercise: abline

You want to illustrate the relation between the variables "Sepal.Length" and "Petal.Width" for each species of the iris data.

  1. Plot the scatterplot between the two variables with one color per species.
  2. Draw the regression lines for each group with the appropriate color.
32 / 81

Text I

You can add text to an existing plot with the function text(). The most important arguments are:

  • x, y: coordinates of the text
  • labels: the text to be displayed
  • pos: the position of the text relative to the coordinate. pos = 0: as is, pos = 1: below, 2: left, 3: top, 4: right.
  • As usual, other graphical parameters apply: cex (size), col, etc.
33 / 81

Text II

plot(5:1, col = "firebrick", pch = 18, xlim = c(0, 6))
text(1, 5, "pos = default")
text(2:5, 4:1, paste0("pos = ", 1:4), pos = 1:4)
34 / 81

Exercise: text

  1. As in the previous exercise, plot the scatterplot between the two variables with one color per species for the variables "Sepal.Length" and "Petal.Width" of the iris data.
  2. Add the Species names in the middle of the points for each species in the right color and with large font.
35 / 81

A functional approach to graphs

36 / 81

From pain we learn

Base R can be surprisingly painful for doing seemingly simple stuff.

Q: What does a programmer do when facing a tedious task?

37 / 81

From pain we learn

Base R can be surprisingly painful for doing seemingly simple stuff.

Q: What does a programmer do when facing a tedious task?

A: S/he automates it!

37 / 81

From pain we learn

Base R can be surprisingly painful for doing seemingly simple stuff.

Q: What does a programmer do when facing a tedious task?

A: S/he automates it!

Base R is so painful, that if you stick to it, it will make you a good programmer (or a masochist!).

Remember though: it's not just painful, it's also extremely powerful!

37 / 81

Hard coding

It's very easy to write code that is specific to your current data! In fact, it's usually the first thing we do, and it works well.

38 / 81

Hard coding

It's very easy to write code that is specific to your current data! In fact, it's usually the first thing we do, and it works well.

plot(iris$Petal.Length, iris$Sepal.Length, 
     col = iris$Species, pch = 20, cex = 2)
text(1.5, 5, "Setosa", 
     font = 2, cex = 4)
text(4, 6, "Versicolor", 
     font = 2, cex = 4, col = 2)
text(6, 7, "Virginica", 
     font = 2, cex = 4, col = 3)
38 / 81

Hard coding: The problem

If your data changes, even slightly, your code is messed up.

Changing Sepal.Length into Sepal.Width loses the legend:

plot(iris$Petal.Length, iris$Sepal.Width, 
     col = iris$Species, pch = 20, cex = 2)
text(1.5, 5, "Setosa", 
     font = 2, cex = 4)
text(4, 6, "Versicolor", 
     font = 2, cex = 4, col = 2)
text(6, 7, "Virginica", 
     font = 2, cex = 4, col = 3)
39 / 81

Hard coding: The problem

If your data changes, even slightly, your code is messed up.

To remember

The data always changes!

40 / 81

Hard coding: The routine

If you want to replicate a hard coded graph to a new data set you:

  1. copy paste the code
  2. change the data
  3. make the adjustments so the graph looks as you wish with the new data
41 / 81
\(\star\): Think of the situation where you advise your boss based on a graph, only to realize a couple of hours later that you simply badly updated a copy-pasted piece of code... and off you go!

Hard coding: The routine

If you want to replicate a hard coded graph to a new data set you:

  1. copy paste the code
  2. change the data
  3. make the adjustments so the graph looks as you wish with the new data

I think I don't need to write that each of these three steps are highly error-prone, and can cost dearly.\(\star\)

41 / 81

Hard coding: The solution

  • very simple: don't hard code!
42 / 81

Hard coding: The solution

  • very simple: don't hard code!

  • OK, here comes some tips

42 / 81
1: Uppercase letters work well to identify parameters.

Tip 1: Define global variables!

  • whenever a variable is repeated twice, use a global variable1 defined at the beginning of the piece of code
43 / 81
1: Uppercase letters work well to identify parameters.

Tip 1: Define global variables!

  • whenever a variable is repeated twice, use a global variable1 defined at the beginning of the piece of code

BAD

plot(iris$Petal.Length, iris$Sepal.Width, 
     col = iris$Species, pch = 20, cex = 2)
text(1.5, 5, "Setosa", 
     font = 2, cex = 4)
text(4, 6, "Versicolor", 
     font = 2, cex = 4, col = 2)
text(6, 7, "Virginica", 
     font = 2, cex = 4, col = 3)

GOOD

FONT = 2
CEX = 4
plot(iris$Petal.Length, iris$Sepal.Width, 
     col = iris$Species, pch = 20, cex = 2)
text(1.5, 5, "Setosa", 
     font = FONT, cex = CEX)
text(4, 6, "Versicolor", 
     font = FONT, cex = CEX, col = 2)
text(6, 7, "Virginica", 
     font = FONT, cex = CEX, col = 3)
43 / 81

Tip 2: Lay bare how you think!

  • when you decide to place some text here, or a legend there, how do you take the decision?

  • you decide based on heuristics (although you may not even notice there was a decision process!)

44 / 81
\(\star\): You can always find an heuristic that led to your decision.

Tip 2: Lay bare how you think!

  • when you decide to place some text here, or a legend there, how do you take the decision?

  • you decide based on heuristics (although you may not even notice there was a decision process!)

  • the game is to extract the (often implicit) rules that made you take a decision\(\star\)

  • if you achieve to make the heuristic explicit: you win since now you can automatize it!

44 / 81

Tip 2: Lay bare how you think!

Remember when I asked to put the names in the middle of the points?

Q: What does in the middle means mathematically?

45 / 81

Tip 2: Lay bare how you think!

Remember when I asked to put the names in the middle of the points?

Q: What does in the middle means mathematically?

A: The barycenter!

45 / 81

Tip 2: Lay bare how you think!

Remember when I asked to put the names in the middle of the points?

Q: What does in the middle means mathematically?

A: The barycenter!

BAD

FONT = 2
CEX = 4
plot(iris$Petal.Length, iris$Sepal.Width, 
     col = iris$Species, pch = 20, cex = 2)
text(1.5, 5, "Setosa", 
     font = FONT, cex = CEX)
text(4, 6, "Versicolor", 
     font = FONT, cex = CEX, col = 2)
text(6, 7, "Virginica", 
     font = FONT, cex = CEX, col = 3)

GOOD

FONT = 2
CEX = 4
plot(iris$Petal.Length, iris$Sepal.Width, 
     col = iris$Species, pch = 20, cex = 2)
bary = aggregate(cbind(Petal.Length, Sepal.Width) ~ Species, 
                 iris, mean)
text(bary[1, 2], bary[1, 3], "Setosa", 
     font = FONT, cex = CEX)
text(bary[2, 2], bary[2, 3], "Versicolor", 
     font = FONT, cex = CEX, col = 2)
text(bary[3, 2], bary[3, 3], "Virginica", 
     font = FONT, cex = CEX, col = 3)
45 / 81

Tip 3: Loop whenever possible!

  • whenever you repeat two statements: use a loop instead!
46 / 81

Tip 3: Loop whenever possible!

  • whenever you repeat two statements: use a loop instead!

BAD

FONT = 2
CEX = 4
plot(iris$Petal.Length, iris$Sepal.Width, 
     col = iris$Species, pch = 20, cex = 2)
bary = aggregate(cbind(Petal.Length, Sepal.Width) ~ Species, 
                 iris, mean)
text(bary[1, 2], bary[1, 3], "Setosa", 
     font = FONT, cex = CEX)
text(bary[2, 2], bary[2, 3], "Versicolor", 
     font = FONT, cex = CEX, col = 2)
text(bary[3, 2], bary[3, 3], "Virginica", 
     font = FONT, cex = CEX, col = 3)

GOOD

FONT = 2
CEX = 4
plot(iris$Petal.Length, iris$Sepal.Width, 
     col = iris$Species, pch = 20, cex = 2)
categ_val = levels(iris$Species)
for(i in seq_along(categ_val)){
    data = iris[iris$Species == categ_val[i], ]
    text(mean(data$Petal.Length), 
         mean(data$Sepal.Width), categ_val[i],
         font = FONT, cex = CEX, col = i)
}
46 / 81

Tip 4: Loop over the tips!

Apply recursively Tip 1, Tip 2 and Tip 3 until you can't any more.

47 / 81

Tip 4: Loop over the tips!

Apply recursively Tip 1, Tip 2 and Tip 3 until you can't any more.

BAD

FONT = 2
CEX = 4
plot(iris$Petal.Length, iris$Sepal.Width, 
     col = iris$Species, pch = 20, cex = 2)
categ = levels(iris$Species)
for(i in seq_along(categ)){
    data = iris[iris$Species == categ[i], ]
    text(mean(data$Petal.Length), 
         mean(data$Sepal.Width), categ[i],
         font = FONT, cex = CEX, col = i)
}

GOOD

FONT = 2
CEX = 4
x = iris$Petal.Length
y = iris$Sepal.Width
categ = iris$Species
plot(x, y, col = categ, pch = 20, cex = 2)
categ_val = levels(categ)
for(i in seq_along(categ_val)){
    who = categ == categ_val[i]
    text(mean(x[who]), mean(y[who]), categ_val[i],
         font = FONT, cex = CEX, col = i)
}
47 / 81

Are the tips useful?

Can those tips be concretely helpful?

To know that, let's summon the copy-paste demon.

48 / 81

Code without tips

plot(iris$Petal.Length, iris$Sepal.Width, 
     col = iris$Species, pch = 20, cex = 2)
text(1.5, 5, "Setosa", 
     font = 2, cex = 4)
text(4, 6, "Versicolor", 
     font = 2, cex = 4, col = 2)
text(6, 7, "Virginica", 
     font = 2, cex = 4, col = 3)
49 / 81

Code without tips: Summoning

50 / 81

Code without tips: Outcome

The demon has immense powers

51 / 81

Code with tips

FONT = 2
CEX = 4
x = iris$Petal.Length
y = iris$Sepal.Width
categ = iris$Species
plot(x, y, col = categ, pch = 20, cex = 2)
categ_val = levels(categ)
for(i in seq_along(categ_val)){
    who = categ == categ_val[i]
    text(mean(x[who]), mean(y[who]), categ_val[i],
         font = FONT, cex = CEX, col = i)
}
52 / 81

Code with tips: Summoning

53 / 81

Code with tips: Outcome

The demon is weak

54 / 81

Tips: Side benefits

If you've followed the tips, guess what:

55 / 81

Tips: Side benefits

If you've followed the tips, guess what:

You can create a function for your graph for free!

55 / 81

Tips: Side benefits

If you've followed the tips, guess what:

You can create a function for your graph for free!

Before

FONT = 2
CEX = 4
x = iris$Petal.Length
y = iris$Sepal.Width
categ = iris$Species
plot(x, y, col = categ, pch = 20, cex = 2)
categ_val = levels(categ)
for(i in seq_along(categ_val)){
    who = categ == categ_val[i]
    text(mean(x[who]), mean(y[who]), 
         categ_val[i],
         font = FONT, cex = CEX, col = i)
}

After

scatter_name = function(x, y, categ, font = 2, cex = 4){
    plot(x, y, col = categ, pch = 20, cex = 2)
    categ_val = levels(categ)
    for(i in seq_along(categ_val)){
        who = categ == categ_val[i]
        text(mean(x[who]), mean(y[who]), categ_val[i],
             font = font, cex = cex, col = i)
    }
}
scatter_name(iris$Petal.Length, 
             iris$Sepal.Width,
             iris$Species)
55 / 81

Why create functions to make graphs?

  1. guards you against, or limits, copy-paste problems
56 / 81

Why create functions to make graphs?

  1. guards you against, or limits, copy-paste problems

  2. facilitates graph replications

56 / 81

Why create functions to make graphs?

  1. guards you against, or limits, copy-paste problems

  2. facilitates graph replications

  3. you don't have to think to implementation details when running the function (reduces mental load)

56 / 81

Why create functions to make graphs?

  1. guards you against, or limits, copy-paste problems

  2. facilitates graph replications

  3. you don't have to think to implementation details when running the function (reduces mental load)

  4. it's very easy to include new features to the functions, and all the calls benefit from it

56 / 81

Mental load

Code telling what you do and not how you do it increases productivity tremendously.

57 / 81

Mental load

Code telling what you do and not how you do it increases productivity tremendously.

FONT = 2
CEX = 4
x = iris$Petal.Length
y = iris$Sepal.Width
categ = iris$Species
plot(x, y, col = categ, 
     pch = 20, cex = 2)
categ_val = levels(categ)
for(i in seq_along(categ_val)){
    who = categ == categ_val[i]
    text(mean(x[who]), mean(y[who]), 
         categ_val[i], col = i,
         font = FONT, cex = CEX)
}
scatter_name(iris$Petal.Length, 
             iris$Sepal.Width,
             iris$Species)
57 / 81
\(\star\): If in doubt, you just have to jump to the function definition and its associated documentation (that you didn't forget to write of course!).

Mental load

Code telling what you do and not how you do it increases productivity tremendously.

FONT = 2
CEX = 4
x = iris$Petal.Length
y = iris$Sepal.Width
categ = iris$Species
plot(x, y, col = categ, 
     pch = 20, cex = 2)
categ_val = levels(categ)
for(i in seq_along(categ_val)){
    who = categ == categ_val[i]
    text(mean(x[who]), mean(y[who]), 
         categ_val[i], col = i,
         font = FONT, cex = CEX)
}
scatter_name(iris$Petal.Length, 
             iris$Sepal.Width,
             iris$Species)

The code on the right will always be easier to understand than the code on the left.\(\star\)

57 / 81

Why not create functions?

  1. you have a presentation in 30 minutes and have to finish that graph
58 / 81
\(\star\): Note that this assumption is in general false. What we think is specific is often but an instance of a more general problem.

Why not create functions?

  1. you have a presentation in 30 minutes and have to finish that graph

  2. you're making a graph that you think will never replicate\(\star\)

58 / 81
\(\star\): Note that this assumption is in general false. What we think is specific is often but an instance of a more general problem.

Why not create functions?

  1. you have a presentation in 30 minutes and have to finish that graph

  2. you're making a graph that you think will never replicate\(\star\)

  3. the graph is really simple (in terms of lines of code!)

58 / 81

Functions: Summary

  • thinking in functions will change the way you code

  • it will clarify your code: it will be easier to understand and share, and less error-prone

  • due to the high fixed costs, 0 marginal cost nature of functions, you'll gain a lot of productivity

59 / 81

Functional programming: Application

Remember the scatterplot with different colors and a linear fit? Let's redo it.

  1. plot the scatterplot between the variables "Sepal.Length" and "Petal.Width" for each species of the iris data, and add a linear fit
60 / 81

Functional programming: Application

Remember the scatterplot with different colors and a linear fit? Let's redo it.

  1. plot the scatterplot between the variables "Sepal.Length" and "Petal.Width" for each species of the iris data, and add a linear fit

  2. use segments() to shorten the fit to the width of the scatterplot

60 / 81

Functional programming: Application

Remember the scatterplot with different colors and a linear fit? Let's redo it.

  1. plot the scatterplot between the variables "Sepal.Length" and "Petal.Width" for each species of the iris data, and add a linear fit

  2. use segments() to shorten the fit to the width of the scatterplot

  3. transform it into a function, with the appropriate arguments

60 / 81

Functional programming: Application

Remember the scatterplot with different colors and a linear fit? Let's redo it.

  1. plot the scatterplot between the variables "Sepal.Length" and "Petal.Width" for each species of the iris data, and add a linear fit

  2. use segments() to shorten the fit to the width of the scatterplot

  3. transform it into a function, with the appropriate arguments

  4. add the argument line_extend giving how much the length of the segment should be extended, in % of the graph width (default is 0)

60 / 81

Functional programming: Application

Remember the scatterplot with different colors and a linear fit? Let's redo it.

  1. plot the scatterplot between the variables "Sepal.Length" and "Petal.Width" for each species of the iris data, and add a linear fit

  2. use segments() to shorten the fit to the width of the scatterplot

  3. transform it into a function, with the appropriate arguments

  4. add the argument line_extend giving how much the length of the segment should be extended, in % of the graph width (default is 0)

  5. control the arguments given by the user

60 / 81

More Base R graphs stuff

61 / 81

Informative content

So far we've seen only data content. But there's much more to make a good graph: all the surrounding information!

  • plot() arguments:
    • xlab, ylab: x/y axis labels
    • sub, main: subtitle and main title
    • axes: whether to draw the axes
    • ann: if FALSE, cleans all x/y labels
  • legend(): adds a legend
  • title(): adds axis labels and titles (close to previous plot arguments)
  • axis(): function to draw the axes of a plot.
  • mathematical formulas
62 / 81

Plot: informative

plot(-1:1, -1:1, xlab = "xlab", ylab = "ylab", main = "main", sub = "sub", type = "n")
text(0, 0, 'plot(-1:1, -1:1, xlab = "xlab", ylab = "ylab", main = "main", sub = "sub")')
63 / 81

Exercise: informative content

Do the scatterplot between variables "Sepal.Length" and "Petal.Width" of the iris data.

  1. Put appropriate axes labels (i.e. add only the name of the variable).
  2. Put the correlation in the title of the plot.
64 / 81

Title

You can add a title after a plot is done with title().

plot(1:5)
title(main = "This is the title", sub = "This is the subtitle")
65 / 81

Title: mtext

Use mtext to add text in the margin of the graph. Can be used to insert a title.

plot(1:5)
mtext("That's a basic graph", side = 3, 
      line = 1, font = 2, adj = 0)
66 / 81

Adding a legend

You can add a legend to clarify the content of a plot. A legend is a piece of information appearing inside the plotting region.

Here are the main arguments:

  • x, y: the location of the legend (top left corner). There exist shorthands! instead you can use "topleft", "right", etc.
  • legend: the content of the legend (a character vector.)
  • pch, lty, col, lwd: the pch, lty, col, lwd associated to the legend vector
  • bty: whether or not to show the legend box ("o" is default, "n" removes it)

Keep in mind that there are many more arguments.

67 / 81

Legend

68 / 81

Legend in the bottom I

How to have a legend in the bottom? Here's some ready-made code.

legend_bottom = function(..., bty = "n"){
    # Original credits to: https://stackoverflow.com/questions/3932038/plot-a-legend-outside-of-the-plotting-area-in-base-graphics/3932558
    op = par(fig = c(0, 1, 0, 1), oma = c(0, 0, 0, 0), 
             mar = c(0, 0, 0, 0), new = TRUE)
    on.exit(par(op))
    plot(0, 0, type = 'n', axes = FALSE, ann = FALSE)
    legend("bottom", horiz = TRUE, bty = bty, ...)
}
69 / 81

Legend in the bottom II

plot(iris$Sepal.Length, iris$Petal.Width, col = iris$Species, pch = 15)
legend_bottom(legend = levels(iris$Species), col = 1:3, pch = 15)
70 / 81

Legend in the bottom II

plot(iris$Sepal.Length, iris$Petal.Width, col = iris$Species, pch = 15)
legend_bottom(legend = levels(iris$Species), col = 1:3, pch = 15)

Oh, yeah the legend ends up being too close to the label... Remember about "setting the stage"? To make it nicer, you'd need to increase the bottom margin beforehand with, e.g., par(mar = c(7, 4, 2, 2)) :-/ That's one of the reasons why ggplot is so much easier to handle.

70 / 81

Axes

You can modify the axes at will.

axis(i) draws the ith axis with, 1: bottom, 2: left, 3: top and 4: right.

plot(1:5, axes = FALSE)
axis(1)
axis(4)
71 / 81

Axis

The function axis has the following main options:

  • side: where to draw the axis (1: bottom, etc, 4:right)
  • at: where the ticks are drawn
  • labels: the labels at the ticks (usually numbers)
  • lwd: line width of the horizontal line (if side is 1 or 3)
  • lwd.ticks: the line width of the ticks. Default is equal to lwd
  • tck: length of the ticks (in fraction of the plotting region)
  • las: orientation of the text

Many other options.

72 / 81

Axis example

plot(1:5, axes = FALSE, ann = FALSE)
box() # draw a simple box
axis(1, 1:4, c("First", "Second", "Third", "Fourth"), cex.axis = .9)
axis(3, 5, "Fifth", lwd.ticks = 2)
axis(2, 1:3, c("One", "Two", "Three"), las = 2)
axis(2, 4, "Four", col.axis = "red")
axis(2, 5, "Five", col.ticks = "blue", lwd = 2)
73 / 81

Mathematical expressions

You can add mathematical expressions in graphics. Advice: for single mathematical formulas, use only function substitute().

Write a formula inside the function substitute() (a bit like in Latex). Some elements that can compose the formula:

  • x[i] for subscript \(x_{i}\)
  • x^i for superscript \(x^{i}\)
  • x %in% y, \(x\in y\)
  • alpha, beta, gamma, etc...
  • paste(x, y, z): juxtapose the three components

See ?plotmath for more details.

74 / 81

Math

75 / 81

substitute

substitute() contains a second argument. It can be used to replace some variables with numbers:

curve(sin(x)*sqrt(x), 1, 10000, log = "x", axes = FALSE, ann = FALSE)
title(main = substitute(sin(x)%*%sqrt(x)))
box() ; axis(2)
for(i in 0:4) axis(1, at = 10**i, substitute(10^p, list(p = i)), 
                   cex.axis = 2)
76 / 81

Graphical parameters

The function par() contains most graphical parameters (there are 72... ).

You can change the graphical parameters directly with par().

par(cex = 2, lwd = 2)
plot(1:5, type = "o")
77 / 81

par

All plots will have these parameters as default. To reinitialize it, you can:

  1. reinitialize it manually: par(cex = 1, lwd = 1)
  2. Make a "save" when you modify it:
op = par(cex = 2, lwd = 2) # save old params
# make the graphs
par(op) # reinitialize it
78 / 81

par: other parameters

Some useful parameters:

  • mar: the margins of the plot: the space between the axes and the edge of the plot. It's a vector of length 4 (1st is bottom, last is right).
  • cex, cex.axis, cex.lab, cex.main, cex.sub: expansion factor for different situations
  • col, col.axis, col.lab, col.main, col.sub.
  • bg: color of the background (default is white)
  • family: font family: can be "serif", "sans" and "mono"
  • las: orientation of text (1: horizontal, 3:vertical)

See ?par for more details.

79 / 81

Multiple graphs

You can combine multiple graphs in one. Simplest way is to use mfrow:

op = par(mfrow = c(2, 2))
for(i in 1:4) curve(sin(x) * x**i, -10, 10, ylab = paste0("sin(x) * x^", i))
par(op) # reset to an unique frame
80 / 81

Conclusion

I'm afraid you won't be able to make nice graphs from this bare bones introduction!

It only brushes the topic but I hope that you could get some insights along the way! And especially that the functional programming approach convinced you!

Cheers!

81 / 81

Outline

  1. Intro
  2. A bit of graphical code in R
  3. A functional approach to graphs
  4. More graphical code in R
2 / 81
Paused

Help

Keyboard shortcuts

↑, ←, Pg Up, k Go to previous slide
↓, →, Pg Dn, Space, j Go to next slide
Home Go to first slide
End Go to last slide
Number + Return Go to specific slide
b / m / f Toggle blackout / mirrored / fullscreen mode
c Clone slideshow
p Toggle presenter mode
t Restart the presentation timer
?, h Toggle this help
Esc Back to slideshow