CBR Demo

1

• CBR Demo
• Realtor Sample
• Weimin Ma

2
Realtor Example

• This is a CBR House Appraisal system
• The "solution" to a case describes the market
  value of a house, based on five problem features
  size, age, local traffic congestion, tax rate,
  and crime rate
• When a problem is presented to the system, the
  single nearest case is retrieved, and a simple
  adaptation based on house size is performed.

3
Realtor Example
4
Introduction of IUCBRF

• Open-source framework developed in Java to
  facilitate the development of case-based
  reasoning systems
• Handle much of the general, domain-independent
  work, freeing developers to consider only
  domain-dependent aspects of the system
• Allow much faster, modular development of the
  systems, to facilitate swapping in and out of
  components.

5
Capabilities

• Providing standard implementation of components,
  with well-defined interfaces for custom
  implementations
• Implementations exist for components such as
  Domain Definition, Case Base Storage, Retrieval,
  Adaptation, Performance Monitoring, Maintenance,
  and GUIs
• Functionalities include

• Problem and solution feature definition
• Similarity metrics and retrieval
• Adaptation technique of retrieved cases
• Hooks for alternative problem-solving approaches
• Judgment of solutions.

6
Design Issue

• All classes in the framework are
  domain-independent, it is the designers
  responsibility to implement some few remaining
  domain-dependent methods
• IUCBRF maintains independence from the domain and
  other components. Here independence means
  implementation details of one component are
  hidden from other components
• Isolating dependent code through template design
  pattern and factory design pattern.

7
Related Components

• Features
• Cases
• Case Base/Indexing
• Domain
• Retrieval
• Adaptation

8
Features

• Primary placeholders for information about both
  problems and solutions
• Following feature types are in the framework

• Double
• Integer
• String
• Boolean
• The set Yes, No
• The set very mild, mild, moderate, severe, very
  severe
• Term vector
• Internet address

9
Cases

• Contains the following Problem, Solution,
  Inactive Contexts, Use Counts, Time Of Creation,
  Source
• Both a problem and solution consist mainly of a
  feature collection
• Inactive contexts describe the contexts of system
  operation (e.g. Normal, Testing) in which a
  case should not be considered
• Use count is incremented each time a case is
  retrieved, and successful use count is
  incremented if the resulting solution is judged
  successful
• Case also know at what point in the systems
  history they were added
• Each case is also associated with a source. For
  example, if a case came from successful problem
  solving episode, then the case source would be
  system-deduced.

10
Case Base/Indexing

• Indexing is the repository in which the cases
  need to be stored
• Three general indexing schemes are implemented in
  the IUCBRF, which are

• Flat case base
• B-tree backed file
• D-tree case base.

11
Domain

• Domain specifies the following Problem and
  solution representations, Problem similarity
  measure, Fallback method
• Problem and solution representations consist of
  the number and type of features, the organization
  of those features, which of those features are
  used as indices, the textual descriptions of each
  feature
• Problem similarity measure is a set of weights
  for a weighted Euclidean distance over the
  problem features
• Fallback method partitioning problem space into
  equivalence classes, each with a prototype
  problem and a set of weights, one per problem
  feature these weights are used in a linear
  combination of the problem features to obtain a
  solution.

12
Retrieval

• The pre-implemented retrieval in IUCBRF is a k
  nearest neighbor (kNN) algorithm, in which, the k
  cases with a problem description nearest to the
  current problem according to a similarity
  criteria are retrieved
• The designer also has the option of saving the
  distances of retrieved cases from the current
  problem, so they may be used in adaptation.

13
Adaptation

• The adaptation portion of the framework
  encompasses the Reuse, Review, and Revise steps
  first obtaining a solution, and perhaps entering
  a cycle of reviewing and revising that solution
• Standard approaches are

• No adaptation return the solution of the first
  retrieved case as the system solution
• Weighted Average assumes the possibility of
  conversion of a solution to a double value, and a
  weighted average of the solution values is
  returned
• Weighted Majority assumes the possibility of
  conversion of a solution to a double value, a
  weighted vote is taken, with the winning value
  being the resulting solution.