Method for adapting a similarity function for identifying misclassified software objects

• US 5,438,676 A
• Filed: 06/04/1993
• Issued: 08/01/1995
• Est. Priority Date: 05/10/1991
• Status: Expired due to Term

• Current Assignee: Siemens Corporate Research Incorporated (Siemens AG)
• Original Assignee: Siemens Corporate Research Incorporated (Siemens AG)
• Inventors: Schwanke, Robert W.
• Primary Examiner(s): Kriess, Kevin A.
• Application Number:
  US08/072,657
• Time in Patent Office: 788 Days
• Field of Search: 395/600, 395/650, 395/700
• US Class Current: 717/120
• CPC Class Codes: G06F 11/3604 Software analysis for verifying properties of programs byte-code verification G06F9/44589
• CPC Class Codes: G06F 18/23 Clustering techniques
• CPC Class Codes: G06F 18/40 Software arrangements specially adapted for pattern recognition, e.g. user interfaces or toolboxes therefor

Abstract

A method of reoptimizing the coefficients of a similarity function coefficient estimation as mavericks are resolved in a maverick analysis comprises computing initial weights for each feature and passing the similarity function to an estimation procedure, along with software objects, their group assignments, a peer parameter K and a confidence parameter N. Receiving as output and using updated values for the coefficients to obtain lists of misclassified and poor-confidence mavericks and placing them in a Current Maverick Set. Presenting the Current Maverick Set to an analyst to determine (1) if the maverick should be deferred and placed in the Deferred Maverick Set; or (2) if the maverick is assigned to a certain group it is removed from the Current Maverick Set and placed in the Firmly Assigned Set; or (3) if the input set of software objects should have certain features added to, or removed from them, or (4) if the similarity function coefficient estimation should be returned to the estimation procedure wherein this time, its inputs are: the original set of software objects less the members of the Deferred Maverick Set and the Current Maverick Set plus the members of the Firmly Assigned Set; the weights of the features and the coefficients previously used, which may be modified if need be; and the modified group assignments. Updated values for the coefficients are received, and when maverick resolution is complete, the reoptimizing stops.

Claims

1. A computer-implemented method for utilizing a similarity function coefficient estimation in a maverick analysis wherein said maverick analysis reoptimizes coefficients associated with said similarity function as mavericks associated with said maverick analysis are resolved, said similarity function receiving a set of software objects, a peer parameter K and a confidence parameter N, wherein said software objects are assigned to a group and are defined by features that encompass at least two software procedures, said method comprising the steps of:

(a) computing initial weights for each feature in accordance with a given criterion for estimating significance of said feature;

(b) creating an Unexplainable Set, initially empty;

(c) creating a Firmly Assigned Set, initially empty;

(d) passing, as parameters, said similarity function and said initial weights for each of said features to an estimation procedure, along with said set of software objects, and a parameter, wherein said parameter defines the number of software objects associated with at least one of said features;

(e) receiving, as output parameters, from said estimation procedure updated values for said coefficients of said similarity function;

(f) using as input parameters, said updated values for said coefficients, said peer parameter K, and said confidence parameter N for said Maverick analysis, to obtain lists of misclassified and poor-confidence mavericks, placing said misclassified and poor-confidence mavericks in a Current Maverick Set;

(g) outputting said Current Maverick Set while, flagging a current maverick that is also in said Firmly Assigned Set;

(h) analyzing said outputted Current Maverick Set, resolving one maverick to provide an approved set indicative of one of the following;

(h.1) said one maverick should be deferred and removed from said Current Maverick Set and/or said Firmly Assigned Set and placed in a Deferred Maverick Set,

(h.2) said one maverick is assigned a group assignment and removed from said Current Maverick Set and/or said Deferred Maverick Set, wherein said one maverick is placed in said Firmly Assigned Set, and said group assignment is updated to be the group named in said input,

(h.3) certain software objects out of said input set of software objects should have features altered therein by said analyst,

(h.4) said similarity function should be returned and said weights of each of said features and said coefficients of said similarity function should be varied if need be and in this case, said estimation procedure is used again, wherein its inputs are;

a subset of said set of software objects, comprising said set of software objects less said Deferred Maverick Set and said Current Maverick Set, plus said Firmly Assigned Set;

said weights of each feature and said coefficients previously used and any modified group assignments as specified in step (h.2);

(i) going back to step (e), whereby said Maverick analysis is complete and said method of reoptimizing coefficients stops.

2. A method for utilizing similarity function coefficient estimation in a maverick analysis in accordance with claim 1, wherein said estimation procedure comprises a Tversky Net Tuning Method.

3. A method for utilizing similarity function coefficient estimation in a maverick analysis in accordance with claim 1, wherein said initial weights for each of said features are computed using Shannon information content, defined as w_f =-log(Probability(f)).

4. A method for utilizing similarity function coefficient estimation in a maverick analysis in accordance with claim 1, wherein said initial weights of each of said features and said coefficients of said similarity function are computed using a form of Feature Ratios with Linking.

5. A method for utilizing similarity function coefficient estimation in a maverick analysis in accordance with claim 1, wherein said similarity function is monotonic, matching, and symmetric.