A classifier testing method and system based on compound metamorphic relationship

Abstract

Embodiments of the present invention provide a classifier testing method and system based on compound transformation relations. Wherein, the method includes: compounding multiple unary metamorphic relationships of the classifier into a composite metamorphic relationship; generating a test case based on the composite metamorphic relationship; testing the classifier based on the composite metamorphic relationship and the test case , and get the test results. In the method and system provided by the embodiments of the present invention, multiple unary metamorphic relationships of the classifier are combined into a composite metamorphic relationship, and then the classifier is tested based on the composite metamorphic relationship, and test results are obtained. The test cost can be reduced and the test efficiency can be improved. It can not only effectively use the characteristics of the classifier itself, such as the uncertainty of the meta-method and the uncertainty of the calculation purpose, but also significantly improve the test effect of the classifier. Effective use of the characteristics of the classifier itself can significantly improve the test efficiency of the classifier, improve the error detection ability of the metamorphic relationship, and reduce the test cost.

Description

technical field [0001] The embodiment of the present invention relates to the field of intelligent software new technologies, and in particular to a classifier testing method and system based on compound metamorphic relations. Background technique [0002] Machine learning is an important research field of artificial intelligence technology. It is one of the core goals of machine learning to construct high-accuracy classification, prediction or clustering machine learning models through machine learning algorithms. As machine learning models are used in finance, autonomous driving and industrial The more and more common applications in control and other fields put forward higher requirements for the safety and reliability of machine learning models. Therefore, machine learning models must be tested and verified to ensure safety and reliability. [0003] The key function of a machine learning model is to identify valuable information from a large amount of data. Data sensitiv...

AI Technical Summary

Problems solved by technology

Therefore, the defect detection ability of the simple unary metamorphic relation classifier is limited, and more efficient testing methods are needed

Images