0:00
[MUSIC]
So, in a bar chart for example, we can use glyphs to represent more than just
the actual value of data elements, we can use glyphs to represent their variants.
So in this case if we have bars, we can put a glyph at the top of each bar
indicating the value, but we can also add information, for
example, standard deviation or variants.
And, in fact, where the mean or median value is indicated and
where the data varies, so these could be 25%, 75% bars.
And we can see in some cases that the distribution of data that gets
aggregated and represented as a single bar,
the actual distribution is skewed in one direction or another.
1:09
But mapping table values to color is a very common thing, and
we're all used to this and seeing weather maps where heat is mapped to color,
as for example, this example here.
Quantitatively mapping color is not necessarily
the most effective way for perceptual accuracy.
We can tell that it's colder up here and warmer down here,
we just might not be able to tell how much colder or how much warmer it is unless
we go to the reference here, and try to make an equivalence on this
legend here showing what the average temperature is for this particular month.
And you can do this discretely as well.
This is an example of continuous variables.
You can do this discretely, and we do this in computer science all the time.
If I have a program that I'm trying to optimize, I might change one parameter
some discrete number of times, and some other parameter some
discrete number of times, and try to find the sweet spot in terms of performance.
The lowest performance here I'm displaying in red,
233 is the optimal value here,
and I can see which parameter of values are giving me worse results.
And these things may jump around quite a bit.
Programs tend to behave non-linearly as you change these parameters
because the systems they run on are quite complicated.
But having these glyphs, basically these numbers represented as color as well,
helps you hone in on the good values.
And this one was enabled just using Microsoft Excel.
The conditional formatting option allows you to
represent values with color as well.
3:00
So you can also use glyphs to do high dimensional visualization.
So in this case this is worlds within worlds,
I've set up horizontal and vertical nominal,
categorical axes and when I'm plotting inside
each of these table cells, is a separate plot, it's a scatter plot.
I just happen to have one data point in the scatter plot, so
I've embedded a two dimensional plot in each element of a two dimensional table.
So that enables you to look at four dimensional data
as basically a two dimensional table of two-dimensional plots.
In this case what I'm doing is I've got a two-dimensional table of glyphs.
Each glyph happens to be a two-dimensional plot.
And so I'm looking in this case at life expectancy, and
I'm looking at the rates of infant mortality inside each year from 2000,
2001, 2002, 2003, 2004.
And then vertically I'm looking at the various continents or regions, and
I'm looking at the life expectancy measured inside each one of those.
So I've got an outer table that's looking at the year and
vertically looking at the region of the world, and
then inside each year and each region of the world,
I'm plotting infant mortality versus life expectancy.
You can see some shifts in the data as we move across different areas of the world.
And you can see some shifts in infant mortality, but
you can see that there aren't as many shifts as you move from year to year.
There's another interesting glyph that you can use called Chernoff faces, and
in this case you're creating a table, some other layout, but
the data points are being plotted with faces, and the attributes of the face
are indicating various measures of some dimension of the data set.
In this case this is 12 state judges as
rated by lawyers and you can map certain features, certain aspects,
of the way the state judges were being measured, for example, the number of
times they were overturned, might change the expression from happy to sad.
And the size of the face may be the number of cases they've seen.
So, you know, this face may look quite large, and this face look
quite small because this state judge has reviewed more cases than has this judge.
And so you can't make value judgments based on the appearance of the faces, but
the appearance of the faces and the expressions that they
have can give you another data channel that you can perceive and remember,
actually quite well as a way of finding interesting aspects of the data,
things in the data that stand out.
John C. HartProfessor of Computer Science
