Seeding/Tagging Estimation of Software Errors: Models and Estimates.
Technical rept. 1 Apr 74-30 Jun 76,
POLYTECHNIC INST OF NEW YORK BROOKLYN DEPT OF ELECTRICAL ENGINEERING AND ELECTROPHYSICS
Pagination or Media Count:
This report concerns itself with seedingtagging estimates of the number of software errors based on the number of errors either inserted deliberately in a program seeded or found by debugging tagged, the number of errors found by a debugger unaware of the first set, and the number of errors appearing in both sets. Estimates from 3 models are discussed. Model 1 assumes all errors are equally open to discovery at all times. Model 2 and 3 assume categories of difficulty exist and that any error which appears can be assigned to the proper category. Model 2 does not assume the relative distribution of errors among categories is known, while Model 3 does. The mean and mean-squared error of a maximum likelihood estimate and a modified maximum likelihood estimate are given for all 3 models. It is shown how these quantities vary with certain relations among the total number of errors, size of tagged or seeded set, and size of accompanying sample set. A procedure for determining optimum values for size of tagged or seeded set and number found by the second debugger is outlined. Finally, multi-trial estimates for parameters are found and compared with single-trial estimates. Author
- Computer Programming and Software