ggplot2 (part 2)

Loading...

En provenance du cours de Johns Hopkins University

Exploratory Data Analysis

4013 notes

Johns Hopkins University

Exploratory Data Analysis

4013 notes

Cours 4 sur 10 dans Specialization Data Science

This course covers the essential exploratory techniques for summarizing data. These techniques are typically applied before formal modeling commences and can help inform the development of more complex statistical models. Exploratory techniques are also important for eliminating or sharpening potential hypotheses about the world that can be addressed by the data. We will cover in detail the plotting systems in R as well as some of the basic principles of constructing data graphics. We will also cover some of the common multivariate statistical techniques used to visualize high-dimensional data.

À partir de la leçon

Week 2

Welcome to Week 2 of Exploratory Data Analysis. This week covers some of the more advanced graphing systems available in R: the Lattice system and the ggplot2 system. While the base graphics system provides many important tools for visualizing data, it was part of the original R system and lacks many features that may be desirable in a plotting system, particularly when visualizing high dimensional data. The Lattice and ggplot2 systems also simplify the laying out of plots making it a much less tedious process.

ggplot2 (part 1)6:25

ggplot2 (part 2)13:53

ggplot2 (part 3)9:46

ggplot2 (part 4)10:37

ggplot2 (part 5)8:10

Rencontrer les enseignants

Roger D. Peng, PhD
Associate Professor, Biostatistics
Bloomberg School of Public Health
Jeff Leek, PhD
Associate Professor, Biostatistics
Bloomberg School of Public Health
Brian Caffo, PhD
Professor, Biostatistics
Bloomberg School of Public Health

So I'm gonna start with a little example dataset here.

This dataset comes with the ggplot2 package,

so you can always just load it up after you install the package.

And so this is the mpg dataset.

It looks at miles per gallon for a variety of different types of cars, and

you can see in this data set from the stir that I did, that there's 234 observations,

so there's 234 different types of cars.

And there are 11 variables that are measured.

In particular, I'm gonna be looking at the displacement variable,

which is an indicator of how large the engine is.

So this is the displ variable.

We're gonna be looking at the number of cylinders and

we're gonna look at the highway mileage, so the hwy variable, and

then we're gonna look at the drv variable which is what kind of drive it is,

is it a four-wheel drive, a front-wheel drive or a rear-wheel drive?

So notice how the factors here are labeled appropriately.

So, for example, the manufacturer variable

is labeled by the manufacturer of the car, Audi, Chevrolet, etc.

And then the drv variable, which indicates the type of the drive, is indicated with

a four for four-wheel drive, f for front-wheel, and r for rear-wheel.

So a very simple plot you can make, I call this the hello world for ggplot,

is to call the qplot function.

On the x-axis I have this displacement variable, which is the kind of engine size

variable, and then on the y-axis I got the highway mileage.

And you can see that I specify the data frame with the data argument, so

I say data equals mpg, so did the data comes from this mpg data frame?

And so that's very simple.

You can see that the plot that it makes looks very different from the traditional

base plot.

You see there's kind of a gray background, and there's white grid lines behind it.

The points are closed, solid circles,

rather than the typical open circles from the base plotting system.

And then there are labels on the x and the y axis.

One of the things we can do is modify some of the aesthetics.

So we can highlight different subgroups of the data.

So one of the subgroups can be determined by, so

there's lots of different cars here.

Some of them are front-wheel drive, some of them are rear-wheel drive, and

some of them are four-wheel drive.

So we can separate those observations out by looking at the drv, the drive variable.

And so I've specified the x and y coordinates just like before.

I specified the data frame just like before.

And then another argument I have is the color variable.

And I'm gonna say that the color is mapped to this drive variable, drv.

And all that says is that the different levels of the drive variable

will be each assigned a different color.

And notice I don't specify what those colors are,

they're specified automatically.

So you can see on this new plot here that the front-wheel drive cars are in green,

the rear-wheel drive cars are in blue, and the four-wheel drive cars are in red.

And so you can see that most of the front-wheel drives tend to have

the highest mileage, the four-wheel drive tends to have the lowest mileage, and

the rear-wheel drive is something in the middle.

And I noticed that the legend was placed on the plot automatically and

the color coding the different levels of the factor variable.

I didn't have to do anything special.

And so it's very nicely organized and thought out, and

you don't have to do anything.

Everything is done automatically.

Another thing that you sometimes want to add is a statistic.

So a statistic is just some summary of the data.

And so the summary that we've chosen to add here is is a kind of smoother or

the more technical name is called low S, and it smooths the data.

So you can see the overall trend in the data set,

and you can see that I do this by adding the geom argument,

and the geoms that I wanna put on this plot are these two types.

One is I wanna add the points themselves, so I can see the data,

and then I wanna add a smooth geom.

And the smooth is this blue line that goes across, and

the 95% confidence intervals for that line are indicated by the gray zone.

You can make a Histogram with the qplot function,

by only specifying a single variable.

So here I only specify the highway variable.

And it shows me the highway mileage for all the cars in the data set.

But then again, I like to kind of separate out

which cars are four-wheel drive, which cars are front-wheel drive, etc.

So again, I specify it.

Instead of the color argument, I specify the fill argument.

Which says the different elements of the Histograms are gonna be filled

with different colors, based on what drive they are.

So, here you can see a similar picture that you saw before,

which is that the four-wheel drive vehicles tend to have the lowest mileage.

And the four-wheel drive tend to have the highest mileage.

Another feature of ggplot is called facets, and

facets are like panels in lattice.

The idea is that you can create separate plots,

which indicate, again, subsets of your data indicated by a factor variable.

And you can make a panel of plots to look at separate subsets together.

So, one option to be to color code the subsets according to different colors,

like we did before.

But if you have a lot of data points, that can be tricky to look at, and then

all of the colors might overlap and it may difficult to see the separate groups.

So an easier way to do that is to say,

split out the three groups into separate panels and make three separate plots.

So that is what we've done here.

On the left side here, I've got three different scatter plots of

the displacement versus the highway mileage, and then

I've split it out by the different drives, so four-wheel drive, front, and rear.

And so you can see the relationship for these three sub groups, and it's just

a different way to look at the data rather than, say, color coding the three groups.

And I specify this with the facets variable.

The fastest variable takes a format that's basically

a variable on the left-hand side and a variable on the right-hand side.

And they're separated by a tilde.

And so the variable on the right-hand side,

determines the columns of the panels.

And the variable on the left-hand side indicates the rows of this kind of

matrix here.

Now notice that in the left plot, which is outlined by the red box,

there's only one row.

And so,

there's no variable that indicates how many rows there should be in this plot.

And so that's why in the facets argument, I have a dot on the left-hand side.

On the right-hand side, I've got the drv variable,

which indicates how many columns.

And now, because there are three levels to this drv variable,

there are gonna be three columns.

And so, since there's only one row,

there's gonna be three plots all in a row like that.

On the right-hand side in this plot that's outlined with the blue box,

I've got three histograms.

And notice that I've put the drv variable on the left-hand

side of the facets argument now, so that indicates I want three separate rows.

Because there's nothing on the right-hand side of the tilde, I have no columns, so

I just have the one column.

No extra columns, I just have the one column for the three plots.

And now I've got the three histograms.

You can look at the highway mileage divided up by the three groups.

So, that's a quick example using the Q plot function,

using some of the data in the ggplot package.

I want to go through a slightly more involved example,

using the data set that comes from here, from Johns Hopkins.

So this comes from the Mouse Allergen and Asthma Cohort Study, which is a study

conducted here at Johns Hopkins of Baltimore children, aged five to 17.

These children all had persistent asthma, with an exacerbation in the past year.

And the overall goal of the study was to study the indoor environment, so the home

environment of these children, and its relationship with asthma morbidity.

So if you're interested in seeing what a little bit of this was about,

we have a recent publication and I give the link here.

So here's a little bit of the MAACS data.

You can see that obviously there are 750 observations and

I just put five variables here for the sake of demonstration.

One is the id variable.

The second one is the eNO.

So, exhaled nitric oxide is a measurement that we take.

Roughly it corresponds to a level of pulmonary inflammation, so

a large value of eNO indicates some pulmonary inflammation.

The secondary variable here that I want to highlight is fine particulate matter.

So this is dust that is less than 2.5 microns in diameter.

So it's very fine dust.

And the last variable I want to point out here is these mouse positive variables.

So this is a skin test that the children do,

which tells us whether they're allergic to mouse allergen or not.

So, here's a basic histogram of exhaled nitric oxide.

So this is the log of exhaled nitric oxide.

And you can see it has kind of an interesting shape.

It looks like there's two peaks or maybe even three peaks there,

as you go across the histogram.

At the bottom of the plot here, I've got the code that makes this histogram,

in case you're wondering how you make it.

And so I use the key plot function to make the histogram.

So now I've made another histogram, but I've color coded the different groups, and

the groups are determined by this mouse positive variable.

So I've separated out the people who are kind of sensitized to mouse allergen and

the people who are not sensitized.

So you can think of the children that are sensitized to mouse allergen are more

allergic and they may be more sensitive to a variety of environmental triggers.

And so you can see that roughly from the data,

the blue bars are slightly higher and the red bars tend to be slightly lower.

So that suggests that the children there are mouse positive have

slightly higher pulmonary inflammation on average.

Another way to visualize this data is to do a density smooth.

So you can see that on the left-hand side,

I add the geom density to this log(eno) variable.

You can see there are at least two peaks from the density smooth, and

if you separate those peaks out, on the right-hand side,

notice I say color equals mopos.

So I split out the colors by whether they're positive to mouse allergen or not.

And you can see that the two peaks roughly correspond to whether they're allergic to

mouse allergen or not.

And so, this is a nice way to visualize this kind of one-dimensional data.

Now, if we want to look at some scatter plots,

I want to see whether exhaled nitric oxide is related to the level of fine

particulate matter in the home, and so I'm gonna look at PM2.5 eNO.

On the very left most side here, I just make a simple scatter plot of log(eno) and

then log(pm25), and it's a little difficult to see exactly

what the relationship might be, but we'll go piece by piece here.

In the middle plot here, I've made the same scatter plot, but

I've separated the two groups, or the mouse positive and the mouse, so

the non-allergic and the allergic children are separated by different shapes.

So here I specified the shape argument, and

I assigned the shape to be this mouse positive variable.

Now, it's not really easy to see, the two groups here.

So there's one ground that's triangles and another group that's circles, and

because of the overlapping of the points, it's a little hard to see.

So in the right-hand side here, instead of separating the two groups by shapes,

I separated them by color, and

it's a little bit easier to see the two different groups in this plot now.

One of the things you can do is to smooth the relationship between a log PM2.5 and

log eNO, and

I wanna look at how this relationship is different in the two groups.

So I set the geom to be a point and a smooth, and

rather than use low S, I'm just gonna use a standard linear regression model.

So I say method equals LM.

And that way I can look at the linear relationship between PM2.5 and

eNO by whether they are allergic to mouse or not.

And you can see that, roughly speaking, in the non-allergic children, so

amongst the red dots here.

There's a little bit of a negative relationship but

it's not particularly strong, if you factor in the confidence intervals there.

And then within the allergic children,

there appears to be an increasing relationship between PM2.5 and eNO

So, another way to look at the same data is to split it out with facets.

So rather than overlapping the two groups and then color-coding them separately,

I can just split them out into two plots using the facets argument.

So here I specify the facets argument and

I say I want two columns, designated by the mouse positive variable.

So no and yes, and so it'll be two columns.

And then I smooth the relationship within each panel,

and you can see that again it's the same story amongst the mouse negative children,

there's a small decreasing relationship.

And amongst the mouse positive children, there's a slight increasing relationship.

So the qplot function is a simple function that you can use to make very quick plots.

And it's analogous to the plot function, so you specify x, y,

you specify your data, and then there's a variety of options that you can choose.

There are a lot of nice built-in features,

so if you wanna color code subsets of the data, it's very easy to do.

If you wanna split out different panels,

that's also very easy to do with the facets.

You can choose different plotting symbols with the shape argument.

And so, it has a lot of nice things that you can do very quickly,

but they are still very powerful.

The syntax of of the function is somewhere in between kind of the base plotting and

the lattice package.

I think the graphics that are produced are very nice,

if you like that type of design.

But if you don't, if there are features you don't like, or

you don't like the design of this particular function.

It's a little bit tricky to kind of modify the qplot function to suit your needs.

If you wanna do a lower level customization of different aspects

of the plot, you really have to go into the guts of the ggplot function, and

that's something that I'll talk about in the next lecture.

Explorer notre catalogue

Rejoignez-nous gratuitement et obtenez des recommendations, des mises à jour et des offres personnalisées.

Coursera propose un accès universel à la meilleure formation au monde, en partenariat avec des universités et des organisations du plus haut niveau, pour proposer des cours en ligne.

© 2018 Coursera Inc. Tous droits réservés.

Télécharger dans l'App Store

Disponible sur Google Play