A measure of the difference between test sets for generating controlled random tests

Abstract

Objectives . The problem of constructing the characteristics of the difference between test sequences is solved. Its relevance for generating controlled random tests and the complexity of finding measures of difference for symbolic tests are substantiated. The limitations of using the Hamming and Damerau – Levenshtein distances to obtain a measure of the difference between test sets are shown.Methods . Based on the characteristic of the interval used in the theory of the chain of successive events, a new measure of the difference between two symbolic test sets is determined. As a difference measure, the distance AD(Ti, Tk) between the test sets Ti and Tk is calculated using the interval characteristic, which is based on determining independent pairs of same (identical) symbols belonging to two sets and calculating the intervals between them.Results. The combinatorial nature of the calculation of the proposed difference measure for symbolic test sets of an arbitrary alphabet and dimension is shown. An example of calculating this measure for various types of test sets, including such as address test sets, is given. Possible modifications are shown and some properties and limitations are determined. The application of the measure of difference is considered for the case of repeated testing of storage devices based on address sequences pA with even p repetition of addresses. For the case p = 2, mathematical relations are given for calculating the intervals and distances AD(Ti, Tk) for address sequences 2A used for controlled random testing of storage devices. The main attention is paid to binary test sets, when the task of calculating given difference metric is reduced to the classical assignment problem using the Hungarian algorithm. The computational complexity of the Hungarian algorithm is estimated by the relation O(n4). As an alternative to the Hungarian algorithm, an algorithm for calculating the considered difference measure is proposed, the complexity of which is much less and has an estimate equal to O(n2). The experimental studies confirm the effectiveness of the proposed algorithm.Conclusion. The proposed difference measure extends the possibilities of generating test sequences when generating controlled random tests. It is shown that test sets, which are indistinguishable when Hamming distance is used as a measure of difference, have different values of AD(Ti, Tk) that allows to make more accurate classification of randomly generated sets as candidates for test sets.