Using affinity measures with supervised classifiers

• US 20050278352A1
• Filed: 05/28/2004
• Published: 12/15/2005
• Est. Priority Date: 05/28/2004
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Bradski, Gary R.
• Granted Patent: US 8,117,143 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 18/22 Matching criteria, e.g. proximity measures
• CPC Class Codes: G06F 18/24323 Tree-organised classifiers

Abstract

A non-binary affinity measure between any two data points for a supervised classifier may be determined. For example, affinity measures may be determined for tree, kernel-based, nearest neighbor-based and neural network supervised classifiers. By providing non-binary affinity measures using supervised classifiers, more information may be provided for clustering, analyzing and, particularly, for visualizing the results of data mining.

Claims

1. A method comprising:

developing a supervised classifier; and

determining a non-binary affinity measure between two data points using said supervised classifier.

2. The method of claim 1 including developing a tree based supervised classifier.

3. The method of claim 2 wherein determining a non-binary affinity measure includes determining whether the two data points would take the same or different branch at each branch point in a pass through a tree taken by both points, accruing one point each time the same branch is taken, and determining the total number of points divided by the total number of branch points visited.

4. The method of claim 2 including developing a forest supervised classifier.

5. The method of claim 4 including summing the affinity between said first and second points in each tree of the forest and averaging the result.

6. The method of claim 1 including determining a kernel based supervised classifier.

7. The method of claim 6 including choosing the number of kernels to equal the number of dimensions in a data space plus one.

8. The method of claim 1 including determining a nearest neighbor based supervised classifier.

9. The method of claim 8 including finding the nearest neighbors to said first point, finding the nearest neighbors to the second point and counting how many of the neighbors have the same class label between the two points.

10. The method of claim 9 including determining the affinity as the ratio of the number of nearest neighbors to each point divided into the number of neighbors that have the same class label between the two points.

11. The method of claim 1 including developing a neural network based supervised classifier.

12. The method of claim 11 wherein determining a non-binary affinity measure includes taking one minus the average absolute value of a node by node difference of 0-1 scaled outputs between the two data points.

13. An article comprising a medium storing instructions that, if executed, enable a processor-based system to:

develop a supervised classifier; and

determine a non-binary affinity measure between data points using said supervised classifier.

14. The article of claim 13 further storing instructions that, if executed, enable a processor-based system to develop a tree based supervised classifier.

15. The article of claim 14 further storing instructions that, if executed, enable the processor-based system to determine whether the two data points would take the same or different branch at each branch point in a pass through a tree taken by both points, accrue one point each time the same branch is taken, and determine the total number of points divided by the total number of branch points visited.

16. The article of claim 14 further storing instructions that, if executed, enable the processor-based system to develop a forest supervised classifier.

17. The article of claim 16 further storing instructions that, if executed, enable the processor-based system to sum the affinity between said first and second points in each tree of the forest and take the average as the result.

18. The article of claim 13 further storing instructions that, if executed, enable the processor-based system to determine a kernel based supervised classifier.

19. The article of claim 18 further storing instructions that, if executed, enable a processor-based system to set the number of kernels to equal the number of dimensions in a data space plus one.

20. The article of claim 13 further storing instructions that, if executed, enable the processor-based system to determine a nearest neighbor based supervised classifier.

21. The article of claim 20 further storing instructions that, if executed, enable the processor-based system to find the nearest neighbors to the first point, find the nearest neighbors to the second point, and count how many of the neighbors have the same class label between the two points.

22. The article of claim 21 further storing instructions that, if executed, enable the processor-based system to determine the affinity as a ratio of the number of nearest neighbors to each point divided into the number of neighbors that have the same class label between the two points.

23. The article of claim 13 further storing instructions that, if executed, enable the processor-based system to develop a neural network based supervised classifier.

24. The article of claim 23 further storing instructions that, if executed, enable the processor-based system to take one minus the average absolute value of a node by node difference of 0-1 scaled outputs between the two data points.

25. A system comprising:

a processor; and

a storage coupled to said processor, the storage storing instructions to develop a supervised classifier and determine a non-binary affinity measure between two data points using the supervised classifier.

26. The system of claim 25 wherein said storage stores instructions to develop a tree based supervised classifier.

27. The system of claim 25 wherein said storage stores instructions to develop a forest supervised classifier.

28. The system of claim 25 wherein said storage stores instructions to determine a kernel based supervised classifier.

29. The system of claim 25 wherein said storage stores instructions to determine a nearest neighbor based supervised classifier.

30. The system of claim 25 wherein said storage stores instructions to develop a neural network based supervised classifier.