Multi-label metamorphic relationship prediction method based on improved rbf neural network

Abstract

The invention discloses a multi-label metamorphic relationship prediction method based on an improved RBF neural network. First, a control flow graph (CFG) is created from the function source code, and then a set of characteristics is extracted from the control flow graph (CFG), and two kinds of features are extracted from the CFG. One type of characteristics is based on node characteristics, and the other is path-based characteristics. The training set is composed of characteristic values ​​and labels, and the improved RBF neural network is used to optimize the calculation of the center of the basis function of the hidden layer RBF neural network. Algorithm—k-means clustering, using AP clustering to automatically find the k value to obtain the number of hidden layer nodes, and construct a Huffman tree to select the initial cluster center, and then construct a label count vector C that reflects information between multiple labels, Then it is linearly multiplied with the RBF neural network basis function center obtained after the improved k-means clustering to obtain a new basis function center, and the RBF neural network model is established to predict the transformation relationship.

Description

technical field [0001] The present invention relates to a method for predicting multi-label metamorphic relationships based on an improved RBF neural network, in particular, a multi-label data set and an improved RBF neural network, that is, an algorithm for obtaining the basis function center of an optimized hidden layer RBF neural network— K-means clustering, use AP clustering to automatically find the k value to obtain the number of nodes in the hidden layer, and construct a Huffman tree to select the initial cluster center, then construct a label count vector C that reflects the information between multiple labels, and then use it with The linear superposition of the RBF neural network basis function center obtained after improving the k-means clustering, obtains the new basis function center, and establishes a multi-label RBF neural network model to predict the transformation relationship, which belongs to the field of software testing. Background technique [0002] Wit...

AI Technical Summary

Problems solved by technology

[0002] With the development of economy and society, more and more software has been developed, but the quality of software has attracted widespread attention. Software testing is an important and indispensable software quality assurance technology, which is used to discover and correct existing problems in software defects and errors, but in most cases the expected output of the program under test is difficult to determine

Metamorphosis testing technology tests the program by checking the relationship between multiple execution results of the program, which can effectively solve the above problems, but the current metamorphosis testing technology also has some deficiencies. The construction method of metamorphic relationship, the lack of effective original test case selection strategy, and the lack of automated testing tools, etc., the test efficiency is low and the cost is high

[0003] At present, the research results on the construction of metamorphic relationships include predicting metamorphic relationships using machine learning technology and machine learning methods using graph kernels. These two methods are single-label metamorphic relationship prediction methods and can only predict one possible metamorphic relationship at a time. Therefore, the present invention proposes a multi-label metamorphic relationship prediction method based on an improved RBF neural network and can predict multiple metamorphic relationships that an application program may satisfy at one time, which can effectively solve the current ubiquitous construction method and lack of practical metamorphic relationships. Improving the construction efficiency of metamorphic relations

Images