env 20065.1 envisioning information lecture 5 – connections ken brodlie [email protected]

30
ENV 2006 5.1 Envisioning Information Lecture 5 – Connections Ken Brodlie [email protected]

Upload: chloe-stone

Post on 28-Mar-2015

213 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.1

Envisioning Information

Lecture 5 – Connections

Ken [email protected]

Page 2: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.2

Outline

• Connections– Applications – networks of information– Graphs and trees– Layout algorithms– Treemaps– Hyperbolic trees

Page 3: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.3

Networks of Information

• In multivariate visualization (lectures 3&4) we are looking among the observations for relations between the variates

• In other applications, we know the structure of the data – for example, the observations might be connected in a graph structure

– eg directory tree

• The challenge is to visualize these connections – sometimes called network visualization

1

2 3

A graph is a set of nodes or vertices (V)and a set of edges (E),

in which an edge joins a pair of vertices.

In a directed graph, the edgeshave an associated direction

Page 4: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.4

Examples of Networks of Information

My Windows2000filestore

Automobile web site- visualizing links

Page 5: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.5

Some More Applications

• Genealogy

• Business management– PERT charts

• Phylogenetic trees showing evolution of life forms

• Can you suggest some more?

http://aleph0.clarku.edu/~djoyce/java/Phyltree/cover.html

http://www.genealogy.com

Page 6: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.6

Online Communities

http://jheer.org/vizster/

Page 7: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.7

Think About …

• Suppose we have a collection of documents…

• .. How could we derive links between these documents?

Page 8: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.8

Drawing Large Graphs

• Layout of large graphs is a challenging problem

• What are the basic principles we want to follow in laying out graphs?

1. ?

2. ?

Page 9: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.9

Force Directed Placement

• Analogy with electrically charged atomic particles leads to a solution

– Too close, repel– Too far apart, attract

• Nodes become particles with forces between them determining their position

• Fruchterman and Reingold developed a famous force-directed placement algorithm

• Suppose d = distance between two nodes, and k = ‘ideal’ distance between nodes

• Attraction force:– fa(d) = d2 / k

• Repulsion force:– fr(d) = - k2 / d

• Ideal distance:– k = C*sqrt(area/number_of_nodes)

C is a constant (found experimentally)Ref: Fruchterman and Reingold,Software Practice and Experience, 1991http://citeseer.ist.psu.edu/fruchterman91graph.html

Page 10: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.10

Forces versus distance

• Notice how combined effect of attractive and repulsive forces works

• Equilibrium at k – does this make sense?

Page 11: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.11

Force-directed Placement Algorithm

• Assign nodes to random positions• Iterate the following

– For each node, calculate force due to repulsion by every other node (fr(d) = - k2 / d)

– For each link, calculate force due to attraction of one node by the other (fa(d) = d2 / k)

– For each node, sum all the attractive and repulsive forces

– Determine displacement by Hooke’s Law for springs: displacement proportional to force

– Limit displacement of any node by:

• Boundary of region

• A maximum step (‘temperature’) which decreases as optimum is reached

Page 12: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.12

Force-directed Placement - Cooling

Rapid initial cooling - quenching Slower subsequent cooling –simmering

Page 13: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.13

Hierarchical Information

• Important special case is where information is hierarchical– Graph structure can be laid out as a tree

http://www.nist.gov/dads/HTML/tree.html

Page 14: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.14

Recursive Algorithm for Binary Trees

• Famous algorithm by Reingold-Tilford (1981)

• Driven by aesthetics including:– Nodes at same level lie along a

straight horizontal line– Left son to left, right son to right– Father centred over sons

• Recursively from bottom-up:– Place roots of left and right

subtrees on top of each other– Push roots apart a certain

distance– Work down levels, pushing any

nodes apart as needed– Draw parent one level above,

midway between children

Page 15: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.15

More general trees

• Extends to more general trees

David Dukehttp://www.comp.leeds.ac.uk/djd/graphs/

Page 16: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.16

• In many tree-drawing applications, it is the size of the node that we wish to visualize– For example, directory structure on hard disc

• Tree maps is a screen filling method which uses a hierarchical partitioning of the screen into regions depending on attribute values

• Alternate partitioning parallel to X and Y axes

Tree Maps

Page 17: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.17

Suppose user hasthree subdirectories:A, B and C

First partition in Xaccording to totalsize of each sub-directory

A B C

Tree Map of Filestore

Page 18: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.18

A B C

Then within eachsubdirectory, wecan partition in Yby the size ofindividual files,or furthersubdirectories

Tree Map of Filestore

Page 19: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.19

Treemap Example

Usenet newsgroups

For history oftreemaps see:www.cs.umd.edu/hcil/treemap-history

Developed over many years by Ben Schneiderman and colleagues

Page 20: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.20

Cushion Treemaps

• Treemaps can be hard to interpret

• Can you decipher the structure here?

• Cushion treemaps is an idea suggested by Jark van Wijk and colleagues

Page 21: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.21

Cushion Tree Maps – The Idea

• Imagine in 1D first, with a flat map• Suppose we have a binary

subdivision• Add a ‘bump’ for each division and

accumulate these recursively• This gives the top curve on right• Use this with a lighting effect to give

the lower image

Page 22: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.22

Cushion Tree Maps – A Filesystem

Download SequoiaView from:http://www.win.tue.nl/sequoiaview/

Page 23: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.23

Hyperbolic Trees

• This is a popular method of displaying hierarchical structures

• Based on ideas of hyperbolic geometry– As Euclidean axioms –

except given a point and a line, there are infinite number of lines passing through the point which are parallel to the line

– See : wikipedia– http://www.geom.uiuc.edu

/java/triangle-area/

• Boundary of disc represents infinity

• Notice how angle of triangle on boundary tends to zero

Triangle in Hyperbolicspace

Page 24: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.24

Visualizing Structure of Web Pages

• Hyperbolic trees have been successfully used to visualize web site structures (Lamping et al, 1995)

• Place home page in centre– with linked pages connected by hyperbolic arcs

– further arcs link to further links

– see: www.acm.org/sigchi/chi95/proceedings/papers/jl_bdy.htm

Escher woodcut… .. inspired:

Page 25: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.25

Hyperbolic Trees

Automobilesweb site

Home pagein centre

Click on linkyou want ...

Page 26: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.26

Hyperbolic Trees

Auto Historymoves to centre of screen

Click on nextlink...

Page 27: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.27

Hyperbolic Trees

Henry Fordis now at the centreand so on...

Page 28: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.28

Hyperbolic Trees

Also worksfor familytrees...

Page 29: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.29

Star Trees

• Now commercially available from Inxight

– Spun out of Xerox PARC– www.inxight.com– Now called ‘star trees’ ;)

Page 30: ENV 20065.1 Envisioning Information Lecture 5 – Connections Ken Brodlie kwb@comp.leeds.ac.uk

ENV 2006 5.30

Further Reading

• Excellent review article:• Graph Visualization and Navigation in Information Visualization:

a Survey (Herman, Melancon, Marshall)• IEEE Trans Vis and Computer Graphics, Vol 6, 2000