visualising multi-objective populations with treemaps

Post on 13-Apr-2017

133 Views

Category:

Data & Analytics

1 Downloads

Preview:

Click to see full reader

TRANSCRIPT

VisualisingMulti-

objectivePopulations

with Treemaps

David Walker

Visualising Multi-objective Populationswith Treemaps

David WalkerCollege of Engineering, Mathematics and Physical Sciences

University of Exeter, UK

VizGEC 2015

VisualisingMulti-

objectivePopulations

with Treemaps

David Walker

Multi-objective Visualisation

Visualising multi-objective populations is an importantaspect of EC

0.0 0.2 0.4 0.6 0.8 1.00.0

0.2

0.4

0.6

0.8

1.0

−1.5 −1.0 −0.5 0.0 0.5 1.0−1.0

−0.5

0.0

0.5

1.0

1.5

f1 f2 f3 f4 f50.0

0.5

1.0

1.5

2.0

VisualisingMulti-

objectivePopulations

with Treemaps

David Walker

Treemaps

Visualise hierarchical data (often used for clustering)

Each node has an associated “value”

The value of a node is represented by the amount of spaceassigned to that node in the treemap

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Goal:

Represent a multi-objectivepopulation using a treemap

VisualisingMulti-

objectivePopulations

with Treemaps

David Walker

Multi-objective Trees

Step 1: Pareto sortingConstruct a partial ordering of individuals using Pareto sorting– this results in a graph

Step 2: Prune edges using dominance distanceRemove edges such that each node has exactly one parent node(retain the parent with the smallest dominance distance) andinsert an artificial “root” node using the global best

A

B

C

D

E

F

nrA

B

C

D

E

F

VisualisingMulti-

objectivePopulations

with Treemaps

David Walker

Multi-objective Treemaps

0 2 4 6 8 10

f1

0

2

4

6

8

10

f 2A

B

C

D

E

F

G

H

I

A

D

I E

F

B

G

H

C

VisualisingMulti-

objectivePopulations

with Treemaps

David Walker

An Example: DTLZ2

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

VisualisingMulti-

objectivePopulations

with Treemaps

David Walker

Enhancing the Treemap

Order the nodes to enhance the clarity of the treemap

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Many-objective Treemaps: 3- and 5-objective DTLZ2

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

VisualisingMulti-

objectivePopulations

with Treemaps

David Walker

Circular Treemaps

In the treemaps presented so far it is difficult to observedominance relationships – instead use circular treemaps

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

A

A

VisualisingMulti-

objectivePopulations

with Treemaps

David Walker

Many-objective Circular Treemaps

3-objective DTLZ2

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

5-objective DTLZ2

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

VisualisingMulti-

objectivePopulations

with Treemaps

David Walker

Revealing Additional Information

Visualising Crowding Distance – as used in NSGA-II

0 2 4 6 8 10 12 14 16

f1

0

2

4

6

8

10

12

14

16

18

f 2

VisualisingMulti-

objectivePopulations

with Treemaps

David Walker

Visualising Objectives

Hughes defined a many-objective radar design problemcomprising 9 objectives, which group into three categories:

1 range objectives

2 velocity objectives

3 transmission time

M3 M7 M5

M9 M4

M8

M6

M2

M1

VisualisingMulti-

objectivePopulations

with Treemaps

David Walker

Summary

Treemaps provide a useful 2-dimensional way of visualisinghierarchical data arising within evolutionary computation

Multi- and many-objective populations

Objectives

Highly scalable

Highly flexible – ordering, colouring and node size can bedefined to suit a particular application

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

M3 M7 M5

M9 M4

M8

M6

M2

M1

VisualisingMulti-

objectivePopulations

with Treemaps

David Walker

Future Work

Expand the range of applications within evolutionarycomputation (genetic programming solutions)

Further refine the treemap layout algorithms to suit theuse of treemaps within evolutionary computation

Identify other means of constructing trees to representmulti-objective populations that are not adversely affectedby treemaps

top related