Visualization method and visualization system

• US 6,873,325 B1
• Filed: 06/30/2000
• Issued: 03/29/2005
• Est. Priority Date: 06/30/1999
• Status: Expired due to Term

• Current Assignee: Bayes Information Technology, Ltd.
• Original Assignee: Bayes Information Technology, Ltd.
• Inventors: Valtonen, Kimmo Antero, Kontkanen, Petri Tapani, Myllymäki, Petri Jukka, Tirri, Henry Rainer, Lahtinen, Jussi Mika Antero, Silander, Tomi Viljam
• Primary Examiner(s): Jankus, Almis R.
• Application Number:
  US10/019,477
• Time in Patent Office: 1,733 Days
• Field of Search: 345/419, 345/440, 345/853, 345/952, 345/953, 345/966, 345/967, 345/968, 345/969, 382/190, 382/159, 707/7, 707/6, 707/101, 706/6, 706/48
• US Class Current: 345/440
• CPC Class Codes: G06F 18/213 Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods

Abstract

A method and system for transforming a multivariate data domain into a low-dimensional visual representation. Probabilistic models of the data domain are utilized, and at least one probabilistic model is used to produce at least one predictive distribution. The predictive distributions are used as inputs to the visualization process, where the multidimensional space is converted to a low-dimensional space. In this process data vectors are considered similar, for example, if the corresponding instances of a predictive distribution, conditioned with the variable value assignments found in the data vectors, are similar. Consequently, similarity is not defined directly using the physical properties of the data vectors, but indirectly through the probabilistic predictive model(s).

Claims

1. Method for generating visual representations of multidimensional data domains, which method comprises the steps of:

selecting data to be visualized from at least one data source, and

choosing the number of dimensions to be used in the visualization, characterized in that the method further comprises the steps of;

constructing a set of probabilistic models,

generating a set of predictive distributions from said set of probabilistic models, and

using at least one predictive distribution belonging to said set of predictive distributions, determining a visual location for each data vector to be visualized.

2. A method according to claim 1, characterized in that it further comprises the step of storing at least one probabilistic model belonging to said set of probabilistic models.

3. A method according to claim 1, characterized in that it further comprises the step of generating a visual representation of the data domain using said determined visual locations.

4. Method according to claim 1, characterized in that in said step of constructing a set of probabilistic models, the model construction is based at least partly on a set of sample data from said at least one data source.

5. Method according to claim 4, characterized in that said set of sample data is a set of data consisting of the data selected in said step of selecting data to be visualized.

6. Method according to claim 4, characterized in that said set of sample data is a subset of the data selected in said step of selecting data to be visualized.

7. Method according to claim 4, characterized in that in said step of selecting data to be visualized, a subset of said set of sample data is selected.

8. Method according to claim 1, characterized in that in said step of constructing a set of probabilistic models, the model construction is based at least partly on knowledge about the problem domain represented as prior distributions.

9. Method according to claim 1, characterized in that in said step of constructing a set of probabilistic models, the model construction is based at least partly on knowledge about the problem domain represented as logical constraints.

10. Method according to claim 1, characterized in that at least one probabilistic model belonging to said set of probabilistic models belongs to the family of models known as Bayesian networks.

11. Method according to claim 1, characterized in that at least one probabilistic model belonging to said set of probabilistic models belongs to the family of mixtures of Bayesian network models.

12. Method according to claim 1, characterized in that it further comprises of step of generating data using at least one probabilistic model belonging to said set of probabilistic models, and in that in said step of selecting data to be visualized, said generated data is selected.

13. Method according to claim 1, characterized in that at least one predictive distribution belonging to said set, of predictive distributions is the conditional distribution for at least one domain attribute.

14. Method according to claim 1, characterized in that at least one predictive distribution belonging to said set of predictive distributions is the conditional distribution for at least one latent attribute.

15. Method according to claim 1, characterized in that at least one predictive distribution belonging to said set of predictive distributions is a combination of the conditional distribution for at least one domain attribute and the conditional distribution for at least one latent attribute.

16. Method according to claim 1, characterized in that the number of dimensions used in the step of generating a visual representation is one.

17. Method according to claim 1, characterized in that the number of dimensions used in the step of generating a visual representation is two.

18. Method according to claim 1, characterized in that the number of dimensions used in the step of generating a visual representation is three.

19. Method according to claim 1, characterized in that in said step of determining the visual locations, said visual locations are determined by pairwise distances between data vectors to be visualized, where the pairwise distances are computed by using at least one predictive distribution belonging to said set of predictive distributions.

20. Method according to claim 19, characterized in that in said step of determining the visual locations, a technique known as Sammon'"'"'s mapping is used.

21. Method according to claim 19, characterized in that said set of predictive distributions comprises a conditional distribution and the pairwise distance between a first data vector and a second data vector is the symmetric Kullback-Leibler-distance between a first instance of the conditional distribution, where the conditional variables are assigned the values present in the first data vector, and a second instance of the conditional distribution, where the conditional variables are assigned the values present in the second data vector.

22. Method according to claim 19, characterized in that said set of predictive distributions comprises a conditional distribution and the pairwise distance between a first data vector and a second data vector is defined using at least the probability that a first random outcome drawn from a first instance of the conditional distribution, where the conditional variables are assigned the values present in the first data vector, is different from a second random outcome drawn from a second instance of the conditional distribution, where the conditional variables are assigned the values present in the second data vector.

23. Method according to claim 19, characterized in that in said step of determining the visual locations, a technique known as Sammon'"'"'s mapping is used.

24. Method according to claim 23, characterized in that said set of probabilistic models comprises a naive Bayes model.

25. Method according to claim 1, characterized in that said set of predictive distributions comprises a first conditional distribution for first domain attribute(s) and a second conditional distribution for second domain attribute(s), and in that in said step of determining the visual locations, said visual locations are determined by pairwise distances between data vectors to be visualized, where the pairwise distances are computed by using at least the first conditional distribution and the second conditional distribution.

26. Method according to claim 25, characterized in that said set of probabilistic models comprises a first probabilistic model and a second probabilistic model, and the first conditional distribution is related to the first probabilistic model and the second conditional distribution is related to the second probabilistic model.

27. Method according to claim 1, characterized in that in said step of determining the visual locations, the visual locations are determined by defining a coordinate system where each dimension represents one component of an instance of a predictive distribution belonging to said set of predictive distributions.

28. Method according to claim 1, characterized in that said set of probabilistic models consists of one probabilistic model.

29. Method according to claim 1, characterized in that said set of predictive distributions consists of one predictive distribution.

30. A visualization system, which comprises means for receiving data to be visualized, characterized in that it further comprises

means for constructing a set of probabilistic models using predetermined probabilistic model structures,

means for generating a set of predictive distributions from said set of probabilistic models,

means for determining, using at least one predictive distribution belonging to said set of predictive distributions, visual locations for data vectors, which constitute at least part of the data to be visualized, and

means for producing a visualization using said visual locations.

31. A visualization system according to claim 30, characterized in that it further comprises means for storing the probabilistic model structures.

32. A visualization system according to claim 30, characterized in that it further comprises means for providing a user interface.

33. A visualization system according to claim 30, characterized in that it further comprises means for displaying said visualization.

34. A visualization system according to claim 30, characterized in that it further comprises means for storing said visualization on a computer-readable medium.

35. A visualization system according to claim 30, characterized in that the means for constructing a set of probabilistic models, the means for generating a set of predictive distributions, the means for determining visual locations and the means for producing a visualization are realized as program instructions stored in a memory medium and in that the visualization system further comprises a processing unit for executing the program instructions.