My name is Ashish Mahabal and we'll be
continuing with R and looking at ggplot.
So, much of this material is taken from
Headley's book by the name ggplot2.
We start with the anatomy of a plot.
Last time we saw a little bit of what
grammar of graphics is, so this shows
graphically how the elements in the
grammar of graphics can be brought about.
So you see at the top, that you have
mapping of variables to aesthetics.
So overall, what happens, is that you have
a data set, a data frame as we saw, and
a certain set of mappings, one or more,
and then there are multiple layers.
And for each of those layer,
you again have some data which is
very likely a subset of
the original data that you have.
Some mapping, and associated with that,
a geom and a stat and
a position where that particular layer or
subset should go.
Then, for the overall plot,
you have got one scale for
mapping, and then you've got a coordinate
system and a faceting specification.
So the coordinate system would take
care of whether you have log scale or
linear scale and so on.
And faceting specification is
where you will know what kind of
categorical variables are being used
to separate the data sets into.
Either sublayers or subplots and so on.
So this overall picture is very good for
allowing you to debug your plot.
So if you are trying to build a plot using
layers and you're not getting what you
want, then you can ask yourself
whether you have the correct geom,
whether you have got the correct stat, and
how they should be interacting with each
other, where should the position be,
whether that is correct, is one of your
coordinate system wrong and son on.
So by just keeping this
figure in front of you,
many times it's fairly easy to debug
what is it that you want to look at.
Se there it is again a summary
of what a plot is made of.
Again, there is a data set, a set of
mappings besides geom set and so on.
Right?
And so, example of scale can be this.
It can be a continuous variable or
a discrete variable.
Remember that, that discrete variables
are the ones that we use for doing
categorization and faceting, but even
the continuous variables, you can convert
them using the as.factor function into
a discrete variable and use that as well.
So, what is it that actually
happens when you make an object?
We saw that you can assign your ggplot
command to an object called b and
not have anything rendered
immediately onto the screen.
So what is it that you can
do with such an object?
If b is the object that you have made,
then you can print it.
And this is something that have
been scored automatically if
you're not inside a loop.
Say if you make this object and
simply say P on the command line.
It's going to do the printing of P.
Then you can save that object.
So if you save it will
write out an image for you.
So, rather than going to the screen,
this is what will allow you
to write it out to the disk.
The other interesting thing is summary,
it is as before or loaded and with
this object b if you simply, say summary
with empty parentheses, then you're going
to see what is hidden in the object and
we'll do that to all the objects.
Similarly, save instead of ggsave
is going to save the object itself,
rather than as a plot.
The object as a R object will be saved to
the disc which you can reload later on.
So let's look at the summary of
the P object that we have defined at
the top here.
Where we are again taking
the crts data set and
simply plotting amplitude versus
standard deviation there and
factoring it into different
colors by the object variable.
So you see that though we have used
only amplitude and standard deviation in
the plot the object knows about all
the variables that are in the data frame.
So all of them are listed here.
It also tells you what is
the size of the data frame.
There are 1619 rows in
total with 26 variables.
And then it tells you what
the mapping is for the X's or
the Y's and
what facetting had been used here.
Now in addition to p, if you were
to do a geom_point, like let's do p
+ geom_point and look for the summary
of that, you'll additionally see what,
the stat_identity is, what the geom_point
is, and what the position_identity is.
In this case we have not specified that
explicitly, so we get mostly NULLs, and
FALSE, and so on.
But all that information
in principle is there.
So that is another useful
thing about the laring aspect.
You can keep on piling up various
layers into an object, and
then those layers become available
to you in various different ways.
So, you can say save p and
file name to save the data file, and
then you can reload it again
using the load command.
And then after that you can
add various layers to that,
so, that's a very convenient thing to do.
This is another way
example of adding layers.
So we use here ggplot, and we are using
a slightly different set here.
The set is called diamonds, and
it is available with the ggplot2 library.
So if you have started ggplot2 library
the data diamonds will allow you
to access these commands.
So here we take that set of diamonds and
use the aesthetics of number of carats
in that and price, and then use a cut for
the colors into the plot there.
And so now if we do geom of
you can add that as a layer.
Notice here that we have used
slightly different syntax.
Rather than saying gm_point,
we say that we want a layer and
then geom equal to point.
Now, this is where you can start combining
different stats in geoms effectively.
So, next what we are saying is that we can
simply plot diamonds with aesthetics of
x equal to carrot, so it's simply
going to use that particular variable.
But then to that p,
to that object that we just defined,
we can add a layer with a separate
geom and a separate stat.
So when you plot a histogram,
what is it that is actually happening?
You bend the data, so
that is what you're, statistics is.
And what do you actually plot?
You plot bars of it, so
that is what your geom is,
and so this pairing is automatically
done from, for histogram.
And then when you say simply p,
that is what is going to appear for
you on the screen.
So here is an example of a histogram
as it comes out of ggplot2.
In this case we have
use a binwidth of 0.3.
We can of course use quite interesting and
sometimes useless stats, or geoms for
a particular data set here.
I am simply listing a few
different geoms like bar,
boxplot, contour, line, point, step, text.
There are a total 29 of these,
and just for
fun here, I have got, used the geom of
polygons on this particular data set.
This is the same dataset that we used in
the last lecture, the CIT six class set.
And then you can see that
it doesn't make sense.
But the shape is roughly the same as what
we had got when we had plotted x versus y.
Similarly, there are 15
different statistics.
I have listed a few here.
The book and
various websites will have more.
We will also have it on the website of
the course, so we can have contours there,
we can have density2d, and
that is what I've used here.
I've used density2d, and I've said that
let use bandwidth that is very small,
only 0.01, and that is why you see that
towards the lower parts of xy area,
you see that there
are very dense contours.
As you go out, there is less density in,
you can see that in the contours there.
So some additional points.
You can have the ggplot
with geom_point as we saw.
When you want to add aesthetics,
you can give that as as
arguments to the geom_point 2.
So, if you say geon.color equals to red,
then the points are going
to be plotted as red.
Similarly, in the layers,
you can have various smoothing things.
So here are shown that for
the point method so
that you can provide method
equal to lm as your layer, and
then smoothing can also be done addition,
in addition to that.
Similarly, you can do, scales and axes.
We have not seen an example of that,
but that's homework.
You can take a look at that.
We are adding that as additional layers.
So when we say geom point, it's going
to take whatever the default scale is.
But in addition, you say that oh,
I want log scale for x,
that's sure enough add that as a layer and
you get that.
And this particular example,
we have added both x and y as log scale.
Then similarly,
you can add plot options, so
you don't have to be
stuck with the defaults.
The defaults are reasonable, but normally
for making plots into publications, so
you want the access and labels et cetera,
to be much bigger, so you can specify all
of that as a additional layer simply
by using the command called opts.
Here, I have specified for instance,
what the title should be and
what the aspect ratio should be.
And you can add various
other commands with that.
So, the homework with respect to this is,
take the dataset that we have provided and
we saw examples using only two variables,
amplitude and standard deviation.
But there are several other
interesting variables.
Try to see if you can use them and
separate them in many different ways.
This, what we are covered here
is fairly basic graphics.
We were not trying to visualize
in any very different or
specific way how to go about things.
We were looking at only what is available.
There will be additional modules in
the course where you'll be looking at
visualization, how you should
be plotting the data, et cetera.
Thank you.