HIPYR Hierarchical and Pyramidal Clustering FEP 27-29 January 2003

1
HIPYRHierarchical and Pyramidal Clustering
FEP27-29 January 2003
2
HIPYR
Objective Build a hierarchy / pyramid on a
set of SO

• from a dissimilarity matrix
• based on the data set
• symbolic clustering clusters are  concepts 

3
HIPYR
Parameters
Hierarchy / Pyramid Dissimilarity matrix / Data
set
Dissimilarity Matrix - single linkage -
complete linkage - average linkage -
diameter
Data set - generality degree - increase of
the generality
4
HIPYR

• Data Set
• selection of variables
• generality measure for modal variables

Rule generation (Y / N)
5
HIPYR
Main options
6
HIPYR

• OUTPUT
• Text file
• List of clusters
• For each cluter intension / extension , height
• Generated rules
• Induced dissimilarity matrix / Quality measure
• Base file
• Graphical output (VPYR)

7
HIPYR

• Present status

Last version of program arrived from Dauphine on
the 29th October Mails exchanged between L.
Marco Cruz and Kutlu Pak Experiments made so far
satisfactory
8
HIPYR

• New developments already implemented
• Aggregation indices (except increase in
  generality)
• Computation of a quality measure
• Output as a base file
• Visualisation of the induced dissimilarity
• Visualisation of the SO describing the clusters

9
HIPYR

• THE CLASS - "Class_4/7"
• AGGREGATION HEIGHT - 0.00652645
• THIS CLASS IS FORMED BY THE UNION OF THE
  SYMBOLIC OBJECT
• "beef tallow" AND THE CLASS "Class_2/7"
• THE CLASS EXTENSION (SYMBOLIC OBJECTS)
• "perilla oil ", "cotton seed oil", "beef
  tallow
• LONG SYMBOLIC OBJECT DESCRIBING THE CLASS
• specific gravity 0.86, 0.937
• freezing point -6, 38
• iodine value 40, 208
• saponification value 188, 199
• Major Fatty acid (Linoleic Acid ,Oleic Acid ,
  Palmitic Acid ,
• Myristic Acid , Searic Acid , Capric Acid )

10
HIPYR

• What remains to be done
• Pruning to be done within VPYR
• Generation of rules VPYR ?
• Hierarchical rules
• Taxonomic variables

11
CLINTCluster InterpretationFEP27-29 January
2003
12
CLINT
Objective to interpret clusters,
defined by categorical single-valued OR
categorical multi-valued variables
cat. single-valued ? partitions cat.
multi-valued ? hierarchies, pyramids
13
CLINT

• INPUT
• Variable defining clusters
• Clusters to be interpreted
• Variables to be used
• Measure for generality of probabilistic
  variables
• Computation of maximum generality

14
CLINT
Main options
15
CLINT
Selection of variables to be used for
interpretation
16
CLINT
Selection of variable defining the clusters
17
CLINT
Selection of categories identifying the clusters
to be interpreted
18
CLINT
Symbolic objects
19
CLINT

• OUTPUT
• Text file, containing
• List of cluster members
• Long description of the clusters
• Contribution measures

Base file
20
CLINT

• What remains to be done
• Determination of short object describing the
  cluster
• Hierarchical rules
• Taxonomic variables