سحر الطايي

Test-oriented development as the central core of fast methods or in the term agile emphasizes the generation of test functions and test classes before writing the code. In these methods, the software design is completed based on trial and error and according to the test results. The multiplicity of tests in such methods can be very time-consuming and in contradiction with the main idea of such methods based on rapid software production. Especially the programs that are written for scientific calculations or machine learning may need a lot of time for each execution. In this regard, a new method has been presented by the supervisor under the title of test capability-based development (TsDD). The purpose of this thesis is to eva‎luate the new method based on testability based on its comparison with the well-known and common method of test-based development. In order to check the level of testability, a model has been presented that eva‎luates the level of testability statically and without the need to execute the code. If needed, the code is rebuilt or in other words refactored so that its testability reaches the appropriate level. In this case, it is ensured that the test will be effective. The noteworthy point here is that there is no specific classification for the types of automatic reconstruction methods and it has not been determined what quality each reconstruction improves, therefore, in this treatise, we have tried to specify reconstruction methods based on testability. Another point here is that the effect of a renovation to improve one quality criterion may be contradictory and cause the reduction of other quality attributes. Therefore, in the development method based on testability, it is suggested that after the end of the test phase, the quality of measurement is measured and the code is rebuilt if necessary. In the statistical analysis conducted in this project, we were able to increase the amount of code coverage in Jason, which is a well-known software tool, from 0.899009 to 0.915025 during the process based on testability. On average, the improvement in testability of three well-known and professional packages has been 12.54, which has improved the test time by 8.96 percent for these three well-known packages.

يادگيري ماشين , توسعه مبتني بر قابليت آزمون

machine learnning , testability driven development

Sahar Altaei

Dr. Saeed Parsa