这这一课程中，我们将学习数据挖掘的基本概念及其基础的方法和应用，然后深入到数据挖掘的子领域——模式发现中，深入学习模式发现的概念、方法，及应用。我们也将介绍基于模式进行分类的方法以及一些模式发现有趣的应用。这一课程将给你提供学习技能和实践的机会，将可扩展的模式发现方法应用在在大体量交易数据上，讨论模式评估指标，以及学习用于挖掘各类不同的模式、序列模式，以及子图模式的方法。

Loading...

来自 University of Illinois at Urbana-Champaign 的课程

数据可视化

553 个评分

这这一课程中，我们将学习数据挖掘的基本概念及其基础的方法和应用，然后深入到数据挖掘的子领域——模式发现中，深入学习模式发现的概念、方法，及应用。我们也将介绍基于模式进行分类的方法以及一些模式发现有趣的应用。这一课程将给你提供学习技能和实践的机会，将可扩展的模式发现方法应用在在大体量交易数据上，讨论模式评估指标，以及学习用于挖掘各类不同的模式、序列模式，以及子图模式的方法。

从本节课中

Week 3: Visualization of Non-Numerical Data

In this week's module, you will learn how to visualize graphs that depict relationships between data items. You'll also plot data using coordinates that are not specifically provided by the data set.

- John C. HartProfessor of Computer Science

Department of Computer Science

[MUSIC]

So when we want to visualize hierarchical relationships,

there's a particularly good method for doing that called a tree map.

So as we visualize graphs, and graphs represent relations between data items.

One relation that we'll want to visualize often is hierarchical relationships,

things like parents and child relationships, ancestry and

descendant relationships.

And so in an undirected graph if you have a tree,

any of these nodes could serve as the root.

But often in layouts we can identify a root node.

And we can look at ancestry and descendants relationships

based on that root node based on the edge distance from a node to that root node.

In a directed graph, we can set up

a more definitive parent child relationship using a directed edge.

And we can have child nodes point towards parent nodes,

or the other way around if we set it up that way.

We can specify a root node that has no parent,

in this directed edge representation.

And we can create hierarchies where a node could have multiple parents, where

we could make them an actual tree, in which case each node has a single parent.

And then we can also speak of a tree depth,

which is basically the number of parents a node has at the bottom level of the tree.

Or the number of descendants, the maximum number of descendants a root node would

have along one chain of these edges.

And so we often need to visualize these trees.

This is an example from genetics of fish, and

the nodes in this case are laid out radially.

And then the tree structure is shown on the inside.

Here's another example from the Tree of Life.

And you can go to this website and zoom in on this image and

see a lot more detail than is shown here.

But this is a layout that shows a hierarchical relationship as well.

One of the methods for

displaying hierarchical relationships that I've found quite useful is treemaps.

Treemaps were introduced by Ben Schneiderman in 1992.

And, they map hierarchical organizations of quantities into areas, so

it's a very visual method for displaying hierarchies based on subset operations.

And so, we have an example of this in a utility called WinDirStat,

which displays a Windows file system.

And it represents the entire file system by this rectangle.

And then that root directory is subdivided into smaller rectangles

representing sub-directories.

And each smaller rectangle is subdivided into smaller rectangles which

are sub-directories of that subdirectory.

All the way down to individual rectangles,

where each of these individual rectangles represents a single file.

And so it gives you this nice broad overview of your file system, but

it also shows you how the file system is divvied up into individual directories and

sub-directories and files.

And so it's a good indication of quantitative views, in this case,

of disk space.

It's telling me how my files and

subdirectories are taking up different amounts of disk space.

And so it's mapping disk space to area.

And I can get a pretty good quantitative measure of that

disk space based on the area relationship of each of these sub-rectangles.

So how do we compute a tree map?

Well, if you're given some hierarchy, some tree.

And for a tree map, it has to be a tree.

It can't just be a regular hierarchy.

Each node has to have a single parent.

You have some value, usually at the bottom level.

For that previous example,

this would be the disk space taken up by each individual file.

And so we first need to do a bottom up procedure that takes our tree nodes, and

for the parent nodes of our tree,

adds up the child nodes in order to assign them a value.

So now each of the parent nodes is equal to the sum of its child nodes.

This parent is equal to 4.

It's 1 plus 2 plus 1.

This parent is equal to 11, because it's 4 plus 4 plus 3.

And finally 16, which is the sum of all of these adds up to 16.

And then we're going to use that parent-child relationship and

the values that it represents in order to subdivide some rectangular region.

In order to display how these small quantities at the bottom are organized,

as portions of this full 16 value at the top, at the root node.

And so the first thing we do is we need to subdivide our rectangle into portions

representing the first set of descendants, 11 and 5.

And so we subdivide the rectangle into a portion that is eleven sixteenths and

five sixteenths.

And so, we subdivide it horizontally with a single vertical division

at eleven-sixteenths of the way from this edge to this edge.

Into a portion that's eleven-sixteenths and five-sixteenths.

Now we have two portions,

and these portions are actually taller than they are wide.

So we're going to subdivide them vertically instead of horizontally for

the next step.

So now we've got this portion here on the left representing 11.

We need to subdivide that into portions representing 4, 4, and 3.

So we subdivide them into a portion that's four-elevenths along the way,

another four-elevenths along the way, and then finally another

three-elevenths along the way, representing each of these three nodes.

And now these are wider than they are tall, so

we're going to subdivide them horizontally instead of vertically at the next step.

And so we do this again, and we subdivide this top node into one-fourth,

two-fourth, and one-fourth portions horizontally.

We do this with all the rest of the nodes,

and we get this tree map of all of the nodes we see here.

And so here I've got 1, 2, 1.

One-fourth, one-half, one-fourth of this rectangle representing four units.

I've got 1, 1, and 2 of this rectangle representing 4 units,

and then 2 and 1 of this rectangle representing 3 units.

All together, those represent 11 units.

And on this side, these 5 units, that were five-sixteenths

of the entire triangle are being represented by 3 and 2.

And I've got a slightly thicker line here differentiating between them.

And then this 3 unit is by individual units here,

and this 2 unit is by individual units here.

And so, we're using area and a hierarchy of area into finer and

finer rectangles in order to divvy up this large rectangle into its component pieces.

You'll notice that for example,

this rectangle here is eleven forty-eighths wide

because it's one-third of eleven-sixteenths.

It's one-third of the eleven-sixteenths on this side of this thick black line.

So one-third times eleven-sixteenths is eleven forty-eighths.

This bottom, the height of this is equal to three-elevenths.

It was three-elevenths of this whole unit.

And so we have three-elevenths times eleven forty-eighths is equal to

one-sixteenth, which is exactly the portion one-sixteenth.

Where is it?

Right here, one-sixteenth of the entire area that we wanted to represent.

And so that gives you a good visual connection between the area

of this tree map layout as a portion of the whole.

And the organization of this node representing one unit of

this root node representing 16 units.

Can also see some of the difficulty in doing ordinal connections between area.

This is 2 units of area, and this is 2 units of area.

If I didn't have these numbers here, it may not be as easy to see that this

rectangle is the same area as this more square region is.

So area's not that good at representing ordinal amounts.

But it is a pretty good tool for representing quantities in general.

So the tree map is a good method for visualizing hierarchies because it relates

the region of that hierarchy to a region on the screen, associating it with area.

That is, we've seen area can be a decent but

not great method for ordinal relationships.

[MUSIC]