Patent US5291007A successfully refreshed from Google Patents.
US5652829A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Feature merit generator
Assignee: International Business Machines Corp
•
Filing: 1994-07-26
•
Granted: 1997-07-29
•
Expires: 2014-07-26
Abstract
Abstract The invention discloses steps to compute a new contextual merit measure for feature variables based on a two-fold discovery. First, a feature variable does not exist by itself in distinguishing classes, and second, what is desired is to obtain its correlation to a class in the context of other feature variables. The invention specifies a merit measure that captures the contextual correlation implicitly.
Inventors
- 👤 Se June Hong
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-10-28 08:47
CPC Classifications
G; G06; G06F; G06F18/00; G06F18/20; G06F18/24; G06F18/243; G06F18/24323
Claims Analysis
Total Claims: 24
Claim 1
Independent Claim
145 words
A program storage device readable by a machine, tangibly embodying a program of instructions executable by the machine to perform method steps for generating a feature variable ranking, said method steps comprising: 1) representing within a data base of the computer a plurality of training examples, wherein a training example comprises a vector of feature variables accompanied by its class value; 2) processing an example of a given class value of interest; 3) generating a distance measure for the given class example of interest by a step of sequentially comparing corresponding elements of the feature variables to corresponding elements in at least one of a counter examples' vector of feature variables, the step of comparing comprising; 4) generating merits to specified feature variables by finding that specified feature variables obligational contribution to a class discrimination in a context of the counter examples' corresponding feature variables.
Claim 2
Dependent Claim
145 words
A program storage device readable by a machine, tangibly embodying a program of instructions executable by the machine to perform method steps for generating a feature variable ranking, said method steps comprising: 1) representing within a data base of the computer a plurality of training examples, wherein a training example comprises a vector of feature variables accompanied by its class value; 2) processing an example of a given class value of interest; 3) generating a distance measure for the given class example of interest by a step of sequentially comparing corresponding elements of the feature variables to corresponding elements in at least one of a counter examples' vector of feature variables, the step of comparing comprising; 4) generating merits to specified feature variables by finding that specified feature variables obligational contribution to a class discrimination in a context of the counter examples' corresponding feature variables.
Claim 3
Dependent Claim
29 words
A method according to claim 1, further comprising a step 5) of incorporation the distance measure in a rule for classifying the given class example to an appropriate class.
Claim 4
Dependent Claim
14 words
A method according to claim 1, wherein a feature variable comprises a symbolic feature.
Claim 5
Dependent Claim
14 words
A method according to claim 1, wherein a feature variable comprises a numeric feature.
Claim 6
Dependent Claim
18 words
A method according to claim 1, wherein a feature variable comprises a symbolic feature and a numeric feature.
Claim 7
Dependent Claim
70 words
A method according to claim 1, comprising a step of generating the distance measure D(i,j) between two examples (i,j) comprising a symbolic feature, by summing the feature, Xk, component distances of the two examples, dk(i,j): ##EQU5## where for symbolic feature, X k , dk(i,j)=0 if the k-th feature values for examples, i and j are the same, i.e. x ki =x kj and d k (i,j)=1 if they are different.
Claim 8
Dependent Claim
32 words
A method according to claim 6, comprising a step of determining a specified number of the nearest counter examples per each example of the given class in terms of said distance, D.
Claim 9
Dependent Claim
33 words
A method according to claim 6, comprising a further step of attributing to each member of said nearest subset of counter examples, a contextual strength that is inverse of the distance measure, 1/D.
Claim 10
Dependent Claim
60 words
A method according to claim 6, wherein each nearest counter example's obligatorily contribution to a merit factor of each feature is accumulated by a step of distributing its contextual strength in proportion to the component feature distance measure as (1/D×(d k /D), this quantity being the class distinguishing obligation of the feature in the context of the two examples concerned.
Claim 11
Dependent Claim
42 words
A method according to claim 1, comprising a step of generating the distance measure D(i,j) between two examples (i,j) comprising a numeric feature and utilizing a ramp function, so that d.sub.k (i,j)=(|x.sub.ki -x.sub.kj |)/t.sub.k if |x.sub.ki -x.sub.kj |≦t.sub.k and d.sub.k (i,j)=1 otherwise.
Claim 12
Dependent Claim
27 words
A method according to claim 10 further comprising a step of obtaining the threshold value, t k , for each numeric feature, for the said ramp function.
Claim 13
Dependent Claim
23 words
A method according to claim 1, further comprising a step for ranking the importance of feature variables in accordance with their distance measure.
Claim 14
Dependent Claim
17 words
A method according to claim 1, further comprising building a decision tree based on said feature ranking.
Claim 15
Dependent Claim
27 words
A method according to claim 1, comprising a step of introducing exogenous knowledge of a relevancy of numeric features vis-a-vis symbolic features, for calibrating the distance measure.
Claim 16
Dependent Claim
21 words
A method according to claim 1, comprising a step of treating multiple classes as a series of separate two class problems.
Claim 17
Dependent Claim
25 words
A method according to claim 1, comprising a step of utilizing said distance measure for gainfully discretizing numerical feature variables, when present, into meaningful intervals.
Claim 18
Dependent Claim
58 words
A method according to claim 16, comprising steps of 1) obtaining a span list defined by the example indices, (i,j), and the contextual separation strength 1/D ij 2 ; 2) specializing the span list for each numeric feature variable; and 3) invoking an interval covering on the specialized span list for the purpose of generating optimal cut points.
Claim 19
Dependent Claim
31 words
A method according to claim 17, comprising a step for manually specifying the number of cuts through an interactive monitoring of the progress in cut-merit-score curve within the interval covering process.
Claim 20
Dependent Claim
24 words
A method according to claim 17, comprising a step of automatically determining the number of cuts corresponding to the knee area of the cut-merit-curve.
Claim 21
Dependent Claim
34 words
A method according to claim 17, comprising further refining steps for the cut points, by freeing the discretized features back to the original numeric values one feature at a time in turn, and re-discretizing.
Claim 22
Dependent Claim
21 words
A method according to claim 20, comprising accommodating extra knowledge of given important cut points for providing optimal additional cut points.
Claim 23
Dependent Claim
29 words
A method according to claim 1, comprising a further step of preprocessing the numerical values of a feature by non-linear scaling for the purpose of flattening the value distribution.
Claim 24
Dependent Claim
35 words
A method according to claim 1, comprising emulating an information theoretic quantity for Contextual Information defined as the weighted sum of the Information between class and a feature given a subset of the other features.
References Cited (Backward Citations)
US4658429A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 2000-07-28 | 2001-01-27 | 2001-07-29 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2004-07-28 | 2005-01-27 | 2005-07-29 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2008-07-28 | 2009-01-27 | 2009-07-29 | $7700.00 | Window Closed / Lapsed |