GUI implementation for Supervised and Unsupervised

1
GUI implementation for Supervised and
Unsupervised SUBDUE System
2
Introduction

• Data Mining is a very vast field, which is
  rapidly developing today. It is the use of
  artificial intelligence to find common,
  interesting, or previously unknown patterns in
  large databases.
• One method for discovering knowledge in
  structural data is the identification of common
  substructures (concepts represented as graphs)
  within the data.

2
3
Introduction (Contd..)

• The SUBDUE system, developed by Cook and Holder
  Cook and Holder, 1999 performs data mining on
  databases represented as graphs, i.e. it
  discovers interesting substructures in structural
  data.
• This project deals with the conversion of a
  textual representation of the data into a
  graphical visualization of the data The input
  data given to the program is the outputs provided
  by the existing SUBDUE.

3
4
The Subdue System

• The SUBDUE system is a data-mining tool that
  discovers interesting substructures in structural
  data
• By compressing previously discovered
  substructures in the data multiple passes of
  SUBDUE produce a hierarchical description of the
  structural regularities in the data.

4
5
Unsupervised Subdue
Example of the input graph to the unsupervised
version of the subdue and the discovered
substructure.
5
6
Unsupervised (Contd..)

• The substructure discovery algorithm used by
  SUBDUE is beam search. A substructure consists of
  a definition and a set of instances. The
  substructure definition is a connected set of
  vertices and edges that define a SUBGRAPH within
  G. An instance of a substructure is a SUBGRAPH of
  G that matches, graph theoretically, to the
  definition. The algorithm for the discovery is
  given below.

6
7
Unsupervised Contd.)

• SUBDUE (Graph, BeamWidth, MaxBest, MaxSubSize,
  Limit)
• ParentList
• ChildList
• BestList
• ProcessedSubs0
• Create a substructure from each unique vertex
  label and its single-vertex instances insert the
  resulting substructures in ParentList
• While ProcessedSubs lt Limit and ParentList is
  not empty do
• While ParentList is not empty do
• Parent RemoveHead (ParentList)
• Extend each instance of Parent in all possible
  ways
• Group the extended instances into Child
  substructures
• For each Child do
• If SizeOf (Child) lt MaxSubSize then
• Evaluate the Child

7
8
Unsupervised Contd.)
Insert Child in ChildList in order by value If
Length (ChildList)gtBeamWidth then Destroy the
substructure at the end of the ChildList Processed
Subs ProcessedSubs 1 Insert Parent in
BestList in order by value If Length (BestList) gt
MaxBest then Destroy the substructure at the end
of the BestList Switch ParentList and
ChildList return BestList
8
9
Supervised

• SUBDUE performs supervised graph based relational
  concept learning. The SUBDUE concept learner
  accepts both a positive graph and a negative
  graph , and evaluates substructures based on
  their compression of the positive graph and lack
  of compression on the negative graph.

9
10
Supervised Contd.

10
11
Supervised (Contd.)

• In the figure above each object has a shape and
  is related to other objects using the binary
  relation on and shape. The discovered
  substructure is as shown in the figure. As seen
  above the SUBDUE discovered substructure is from
  the positive example and not from the negative
  example. For this example the best substructure,
  which gives the maximum compression, is triangle
  on a square.

11
12
Subdue - Chemical Compounds

• A DNA sequence can be represented as a very
  simple linear graph, and higher-level
  relationships between different parts of a
  sequence can be mapped to additional edges in
  graph.
• SUBDUE system discovers patterns in the input
  graph in polynomial time.
• SUBDUE system is capable of discovering known
  patterns in the DNA sequence of yeast,as well as
  patterns in yeast DNA that are known to be
  important in other organism, but which have not
  yet been shown to play a role in yeast.

12
13
Subdue - Chemical Compounds

• A DNA sequence can be represented as a very
  simple linear graph, and higher-level
  relationships between different parts of a
  sequence can be mapped to additional edges in
  graph.
• SUBDUE system discovers patterns in the input
  graph in polynomial time.
• SUBDUE system is capable of discovering known
  patterns in the DNA sequence of yeast,as well as
  patterns in yeast DNA that are known to be
  important in other organism, but which have not
  yet been shown to play a role in yeast.

13
14
Subdue - Chemical Compounds

• The figure shown above is the backbone
  representation which gives more meaningful graphs
  then the linear representation.
• This representation separated the base names (A,
  C, T, G) from the vertices representing
  themselves.
• The backbone representation mimics the actual
  chemical structure of the DNA molecule, in which
  the DNA bases are connected by deoxyribose sugars
  to a linear phosphate backbone.

14
15
GUI Design

• Requirements
• The requirements of the GUI are as follows.
• File Dialog Boxes should be added for better user
  access to the input files.
• The entire visual representation of the graphs
  needs to be shown on the screen. Sometimes these
  representations exceed the length of the screen.
  To accommodate these large graphs, scrollbars
  need to be incorporated into the design.

15
16
GUI Design

• User interfaces must be provided so that the user
  can interact with the GUI for displaying the
  results of each and every iteration. So a button
  called Next Iteration, which activates the
  display of substructures on the screen, needs to
  be incorporated.
• A Button called the Compress button should be
  provided to the GUI. This button enables the user
  to see the compressed graph.

16
17
GUI Design

• For the supervised version of SUBDUE, both
  negative and positive graphs need to be
  displayed.
• The vertices of the graph should display their
  labels inside the vertices.
• Since directed edges are used, the arrows with
  appropriate directions should be displayed.
• The language to be used for implementation should
  be portable and be able to run from a browser. It
  should also have good GUI components. So JDK 1.2
  was used to implement the program.

17
18
Implemenation

• Unsupervised SUBDUE
• Unsupervised SUBDUE GUI requires two input files.
  A position file which determines the graph
  position and the output file from the SUBDUE
  which is parsed as per the requirements of the
  GUI by the conversion program.
• The driver class of the applet is the
  unsupervised class. This class initializes the
  applet. When user clicks the next iteration
  button canvas2 class is invoked and best
  substructures found in that iteration is
  displayed. These substructures are arranged by
  their MDL value.

18
19
Implemenation

• Unsupervised SUBDUE
• When user clicks the compress button then canvas1
  class is invoked. This class compresses the input
  graph by replacing the instances of the best
  substructure of the iteration by single vertices.
  The compressed graph will be further compressed
  when the using the results of that
  iteration.when-compressed button is clicked. The
  flow diagrams are as shown below.

19
20
Implemenation

• Supervised GUI Implemenation
• Supervised SUBDUE GUI requires three input
  files.Two position files which determines the
  positive graph position and the negative graph
  position and the output file from the SUBDUE
  which is parsed as per the requirements of the
  GUI by the conversion program.
• The driver class of the applet is the supervised
  class. This class initializes the applet. When
  user clicks the next iteration button canvas2
  class is invoked and best substructures found in
  that iteration is displayed. These substructures
  are arranged by their MDL value.

20
21
Implemenation

• Supervised GUI Implemenation
• When user clicks the compress button then canvas1
  class is invoked. This class compresses the input
  graph by replacing the instances of the best
  substructure of the iteration by single vertices.
  The compressed graph will be further compressed
  when the using the results of that
  iteration.when-compressed button is clicked. The
  flow diagrams are as shown below.

21
22
Implemenation

• Parser for the GUI input.
• The Conversion program parses the output from the
  subdue to make it compatible to the GUI program.
  This can parse both supervised and unsupervised
  GUI.The conversion program takes in longer
  phrases and replaces it by shorter ones It
  eliminates blank lines and arranges the output in
  the format required for the GUI input.
• Each line starts with a small phrase for E.g.
  It(iteration number ), C(for compression),
  v(vertices),e(edges),Val(Value), It(instances).

22
23
Implementation

• Parser for the GUI input.
• The phrase Best substructures indicates the
  starting of the iteration and the phrase Graph
  is compressed using best substructure. Indicates
  a compressed graph information

23
24
Demo Of the Project
24