R Graphics with Ggplot2: Day 2 November 16, 2016

Transcription

1 R Graphics with Ggplot2: Day 2 November 16, 16 Saving your plot You can save you plot with ggsave. ggsave("cars.pdf") ggsave("cars.tiff") The default is to save the last plot created in the format determined by the file extension at size of the current graphics device. You can adjust all these in the ggsave command. Look at the help file. Grouping In ggplot, group is another aesthetic. However any aesthetic that creates distinction between items will cause groups to occur: library(nlme) ggplot(oxboys,aes(age,height)) + geom_point() + geom_line(aes(group=subject)) height age ggplot(oxboys,aes(age,height)) + geom_point() + geom_line(aes(colour=subject)) 1

2 height Subject age ggplot(oxboys,aes(age,height)) + geom_point() + geom_line(aes(linetype=subject)) 2

3 height Subject age Once the grouping is defined, it is in effect for all subsequent layers: ggplot(mpg, aes(displ,, colour=drv)) + geom_point() + geom_smooth() 3

4 35 25 drv 4 f r displ We can add an overall best-fit line by redefining the group in the aesthetic for geom_smooth : ggplot(mpg, aes(displ,, colour=drv)) + geom_point() + geom_smooth(aes(group=1)) 4

5 35 25 drv 4 f r displ Exercise: Using the gapminder dataset (library(gapminder)), make a line graph of life expectancy of each country over time. Colour countries by the continent. Faceting Rather than putting a lot of information in a single graphic, we can split the graphic by certain features and plot a matrix of graphics to see the effect of the feature on the data. This is called faceting. There are two different facet commands, facet_grid and facet_wrap. facet_grid is typically used when you have at least 2 factors and you want the layout to be a matrix with one factor defining the rows and the other defining the columns. p0 <- filter(mpg, cyl!= 5, drv %in% c("4", "f")) %>% ggplot(aes(,hwy)) p1 <- p0 + geom_point() p1 + facet_grid(drv ~ cyl) 5

6 hwy 40 4 f p1 + facet_grid(drv ~.) 6

7 40 hwy 40 4 f p1 + facet_grid(. ~ drv)

8 4 f 40 hwy facet_wrap is typically used when you have a single factor with too many levels to fit in a single row or column p1 + facet_wrap( ~ cyl, ncol=2) 8

9 hwy p1 + facet_wrap( ~ drv + cyl) 9

10 hwy f 4 f 6 f Exercise: Use a histogram to explore the distribution of engine size for different vehicle classes. The default for faceting is to use the same scale for each plot. This is usually a wise choice but you can allow each plot to have its own scale. p1 + facet_wrap(~cyl,scales="free") 10

11 hwy p1 + facet_wrap(~cyl,scales="free_x") 11

12 hwy Facets also work well with groups: p0 + geom_point(aes(color=drv)) + facet_wrap(~cyl, ncol=2) 12

13 hwy 40 8 drv 4 f Note: You cannot use a continuous variable to facet. You must first convert it to a discrete variable. Changing title and axis labels You can add a title or change the axis labels ggplot(economics, aes(date, unemploy)) + geom_line() + xlab('date') + ylab('unemployed (in 1000s)') + ggtitle('unemployment over time') 13

14 Unemployment over time Unemployed (in 1000s) You can add them all with labs() Date ggplot(economics, aes(date, unemploy)) + geom_line() + labs(x = 'Date', y = 'Unemployed (in 1000s)', title = 'Unemployment over time') 14

15 Unemployment over time Unemployed (in 1000s) Date We can modify the axis limits in a similar fashion. p1 <- ggplot(mpg, aes(displ, )) + geom_point() p1 p1 + xlim(-1,4) + ylim(-5,25) Scales in general Labelling and axis limits are really part of scales but quite often they are all you want to change so ggplot2 contains special functions to control just those elements. Scales control the mapping from data to aesthetics. They take your data and turn it into something that you can see, like size, colour, position or shape. Scales also provide the guides like axes marks and legends to allow you to read observations from the plot and map them back to their original values. Every plot created by ggplot has a scale for every aesthetic it uses, even if only implicitly. The default scales will contain reasonable defaults, but you can take as much control as you want by providing your own scale. This is done by simply adding another layer to the plot, like this: p1 + scale_x_continuous("engine displacement") + scale_y_continuous("city mileage") 15

16 40 City mileage Engine displacement All scaling functions are of the form scale_ + aesthetic + _ + name of scale, such as scale_x_continous above. Their first argument is always name, which produces the label for the axes and title for the legend. (Setting just this in the last example was equivalent to using xlab and ylab. But you can also control which values appear in the axes as the tick marks and keys in the legend, using the breaks argument, as well as their labelling using the labels argument. (Notice the parallels between axes and legends.) Here s an example of changing the scale for the axes: p1 <- ggplot(sample_n(diamonds, 100), aes(carat,price, colour=cut)) + geom_point() p1 16

17 15000 price cut Fair Good Very Good Premium Ideal carat p1 + scale_x_continuous(name="carat", limits=c(.2,4), trans="log", breaks=c(.25,.5,1,2,4)) + scale_y_continuous(trans="log10", breaks=1000*c(.5,1,2,4,8,16), minor_breaks=null) + scale_colour_discrete("cut") 17

18 price Cut Fair Good Very Good Premium Ideal Carat (Use minor_breaks to control where to place the fainter grid lines that go between the ticks, or NULL to disable them entirely.) Since log and sqrt transformations are so common, ggplot has defined convenience scale functions for these transformations. p1 + scale_x_log10(breaks=c(.25,.5,1,2,4)) + scale_y_log10(breaks=1000*c(.5,1,2,4,8,16)) 18

19 price cut Fair Good Very Good Premium Ideal carat Lets change the scale for colours. p1 <- ggplot(mpg,aes(,hwy, colour=factor(year))) + geom_point() p1 19

20 40 hwy factor(year) p1 + scale_colour_manual(name="year",values=c("red","blue"))

21 40 hwy Year You can even choose to display in the legend only some of the keys: p1 + geom_point(aes(colour=factor(year))) + scale_colour_manual(name="",values=c("red","blue"),breaks=1999,labels="old Data") 21

22 40 hwy Old Data Exercise: Using the msleep data frame, plot the length of the sleep cycle versus the total amount of sleep. Indicate the animal s diet by colour. Add appropriate labels to the axes and colour legend, and make the colour key labels more reader-friendly. We can further control the placement of the legend (or even turn it off) with the legend.position argument of the theme function: p1 + theme(legend.position='top') 22

23 factor(year) hwy p1 + theme(legend.position="none") 23

24 40 hwy Even finer control of the legend is available through the guides p1 + guides(colour=guide_legend(label.position="bottom",ncol=2)) 24

25 40 hwy factor(year) The default colour scale for continuous variables is scale_colour_hue(). You can use this scale to adjust the gradient used to colour from low to high values of the variable. If you do not like the default colours but don t want to choose your own with scale_colour_manual (probably a good idea), you can use scale_colour_brewer which uses a preselected set of colours that work well together. (Note: this requires that the package RColorBrewer be installed.) These are the options available library(rcolorbrewer) display.brewer.all() 25

26 YlOrRd YlGnBu YlOrBr Purples RdPu Reds YlGn PuBuGn PuRd Oranges OrRd Greens Greys GnBu BuGn BuPu Blues Pastel1 Pastel2 Set1 Set2 Set3 Paired Accent Dark2 Spectral RdYlGn RdYlBu RdGy RdBu PRGn PuOr BrBG PiYG Examples for points: p1 <- ggplot(msleep,aes(brainwt,bodywt,colour=vore)) + geom_point() + scale_y_log10() + scale_x_log10() p1 + scale_colour_brewer(palette="set1") ## Warning: Removed 27 rows containing missing values (geom_point). 26

27 1e+03 bodywt 1e+01 vore carni herbi insecti omni 1e brainwt p1 + scale_colour_brewer(palette="set2") ## Warning: Removed 27 rows containing missing values (geom_point). 27

28 1e+03 bodywt 1e+01 vore carni herbi insecti omni 1e brainwt p1 + scale_colour_brewer(palette="pastel1") ## Warning: Removed 27 rows containing missing values (geom_point). 28

29 1e+03 bodywt 1e+01 vore carni herbi insecti omni 1e 01 Examples for areas brainwt p2 <- ggplot(msleep,aes(brainwt,fill=vore)) + geom_histogram(binwidth=1) + scale_x_log10() p2 + scale_fill_brewer(palette="set1") ## Warning: Removed 27 rows containing non-finite values (stat_bin). 29

30 15 count 10 vore carni herbi insecti omni 5 0 p2 + scale_fill_brewer(palette="set2") 1e 03 1e 01 1e+01 brainwt ## Warning: Removed 27 rows containing non-finite values (stat_bin).

31 15 count 10 vore carni herbi insecti omni 5 0 p2 + scale_fill_brewer(palette="pastel1") 1e 03 1e 01 1e+01 brainwt ## Warning: Removed 27 rows containing non-finite values (stat_bin). 31

32 15 count 10 vore carni herbi insecti omni 5 0 1e 03 1e 01 1e+01 brainwt Coordinate systems Sometimes you want to force certain features on the coordinate system in a graphic. This can be accomplished by the coord_xxx set of functions. We can flip the x and y axis. In fact this is the only way to get horizontal boxplots. library(gapminder) p1 <- ggplot(mpg, aes(class, )) + geom_boxplot() + facet_wrap("year") p1 + coord_flip() 32

33 suv subcompact pickup class minivan midsize compact 2seater Notice that flipping the co-ordinates is not always the same as redoing the graph with the variable roles reversed. ggplot(mpg, aes(displ, )) + geom_point() + geom_smooth(se=false) + coord_flip() 33

34 7 6 5 displ ggplot(mpg, aes(, displ)) + geom_point() + geom_smooth(se=false) 34

35 7 6 5 displ We can force a specific aspect ratio between the x and y axis ggplot(mpg,aes(,hwy)) + geom_point() + coord_fixed() 35

36 40 hwy Positioning Position adjustments apply minor tweaks to the position of elements within a layer. They are particularly useful for bar-type plots, which by default stack the different groups: p1 <- ggplot(diamonds,aes(color,fill=cut)) p1 + geom_bar() 36

37 9000 count cut Fair Good Very Good Premium Ideal 0 D E F G H I J color With argument position = "dodge", you can have ggplot place the grouped bars side-by-side: p1 + geom_bar(position="dodge") 37

38 count cut Fair Good Very Good Premium Ideal D E F G H I J color Point geoms use different position adjustments, primarily to reduce overplotting by slightly moving the marks. Jitter moves the points randomly: ggplot(mpg, aes(class, )) + geom_point() 38

39 seater compact midsize minivan pickup subcompact suv class ggplot(mpg, aes(class, )) + geom_point(position = 'jitter') 39

40 10 2seater compact midsize minivan pickup subcompact suv class The default amount of jitter is quite large, but you can control it with width and height arguments to jitter. {r} ggplot(mpg, aes(class, )) + geom_point(position=position_jitter(width=.3, height=0)) This is getting kind of verbose, so there is a convenience geom_jitter : ggplot(mpg, aes(class, )) + geom_jitter(width=.3, height=0) 40

41 seater compact midsize minivan pickup subcompact suv class Adding text to a plot You can add text to any plot using geom_text, with the text to display in the label aesthetic: ggplot(mpg, aes(, hwy, label = model)) + geom_text() 41

42 jetta new beetl 40 new beetle hwy corolla civic civic corolla civic corolla civic altima civicivic camry a4 camry sonata solara altima solara sonata a4 malibucorolla passat malibu a4 tiburon new camry new beetle camry altima civic jetta a4 gti beetle solara jetta gticivic grand sonata camry a4 prix tiburon quattro grand sonata a4 prix impreza malibu forester altima quattro tiburon camry camry awd solara grand corvette a4 mustang camry camry passat malibu a4 grand prix maxima sonata impreza malibu camry quattro a4 tiburon forester altima solara solara new beetle passat passatpassat new jetta prix gti beetle impreza awd awd awd a4 corvette a4 grand quattro a4 a6 quattro forester quattro prix mustang impreza forester maxima awd awd awd a6 corvette quattro mustang tiburon caravan tiburon forester gti 2wdawd mustang corvette caravan a6 jetta quattro forester jetta 2wd awd caravan mustang grand toyota 2wd cherokee tacoma 2wd 4wd caravan mustang 2wd c1500 pathfinder grand toyota mustang suburban toyota cherokee 4runner tacoma 4wd tacoma 4wd 2wd 4wd 4wd dakota k1500 dakota pickup tahoe pickup durango dakota pickup 4wd dakota expedition c1500 caravan ram expedition mountaineer grand explorer grand toyota cherokee explorer 4runner c1500 dakota pickup suburban durango dakota k1500 4wd 4wd tacoma 4wd 4wd 4wd land pathfinder grand navigator range cruiser cherokee toyota rover 2wd 4wd wagon tacoma 4wd 4wd 4wd navigator mountaineer f150 grand explorer f150 toyota 2wd 4runner pickup pathfinder cherokee pickup tahoe 4wd 2wd tacoma 4wd 2wd 4wd 4wd ram ram navigator durango f150 pickup 4wd 2wd pickup 4wd 4wd and ram c1500 dakota k1500 f150 cruiser durango range suburban pickup tahoe wagon rover 4wd 2wd 4wd grand k1500 cherokee tahoe 4wd 500 ta rango cherokee pickup 4wd 4wd This geom accepts many additional aesthetics that affect the way the text is printed. Look at the help documentation. Clearly, the text geom should be used judiciously, because it quickly becomes unreadable because of overlap. You can use the nudge_x and nudge_y arguments to geom_text to manually adjust the position (as well as the hjust and vjust aesthetics for alignment relative to the x-y position), but when there are too many points there won t be any good way to label them all. To label specific points, just give a custom data argument to geom_text. For instance, to label just the smallest engines: p1 <- ggplot(mpg, aes(, hwy)) + geom_point() p1 + geom_text(aes(label=paste(manufacturer,model,sep="\n")), data = filter(mpg, displ == min(displ))) 42

43 40 hwy honda honda civic civic honda civic honda civic To add just a one-off text annotation, you can also use the convenience annotate function. p1 + annotate('text', 15, 40, label="there is a clear linear trend\nbetween highway and city fuel econom 43

44 40 There is a clear linear trend between highway and city fuel economy hwy The label geom puts the text inside a rectangle for a better visibility, which is particularly handy for this purpose, if also (currently) considerably slower than the text geom: p1 + annotate('label', 15, 40, label="there is a clear linear trend\nbetween highway and city fuel econo 44

45 40 There is a clear linear trend between highway and city fuel economy hwy ** Exercise: Using msleep data, plot total sleep against body weight using the species name instead of a point. (Use a log axis if warranted.) ** 45