We have almost entered in an era where applications monitor our lives. Starting from a primary BP machine to conventional angiography, we depend...Read More
Regression testing is complex when the interface is reformed. It is possible that many of the current test cases cannot be performed with the new interface. This can bound automation in regression testing and therefore considerably increase the cost of regression testing. That’s where artificial intelligence can be introduced. The regression test cases that are affected by the change in structure are automatically regenerated and interface structure is modeled, where possible. It is extremely useful to have knowledge regarding where faults are likely to occur while organizing testing, this can be fed into decisions regarding how much testing effort is consumed on each module. AI (Artificial intelligence) techniques can be used to calculate which modules are likely to be specifically susceptible to faults.
While this blog covers the use of different AI techniques within the different phases of software testing, I was a little shocked that there was no material on one of the most important applications of AI techniques to software testing, the use of metaheuristic algorithms in test case generation. Despite the lack of material on using metaheuristic algorithms, it has the power to change software testing considerably.
When manually making a test suite, the tester is more absorbed on how to test a function. By using a planning system, the tester can focus on what function to test and the path is taken care of. An extra benefit of this is that a planning system is not limited in any way when creating the path and may regularly find a path that was never expected by the tester. This problem is a very important one to battle.
Another technique of creating GUI test cases fakes a beginner. An expert user of a system inclines to follow a more predictable and direct path through a GUI, whereas a beginner would follow a more random path. A beginner is then likely to search more probable states of the GUI than an expert.
The difficulty lies in creating test suites that fake novice system usage. Genetic algorithms have been proposed to solve this problem. Beginner paths through the system are not random paths. First, a Beginner will learn over time and usually won’t make the same mistakes continually and secondly, a beginner is following a plan and possibly has some system or domain knowledge.
Genetic algorithms work as – a group of genes are formed randomly and then are subjected to some task. The genes that don’t complete the task best are the ones which are discarded and those which complete the task best are kept. The process is again repeated with the surviving genes being duplicated and the rest of the set packed in with more random genes. In the end one gene (or a small set of genes if there is some threshold group) will be the only gene in the set and is naturally the best fit for the given problem.
In the case of GUI testing, the method works as above – Each gene is basically a list of random integer values of some fixed length. Each of these genes signifies a path through the GUI. For example, for a given tree of widgets, the first value in the gene (which will be called as an allele) would select the widget to activate on, the following alleles would then fill in input to the widget liable on the number of possible inputs to the widget (for example: a pull down list box would have one input; the selected list value). The success of the genes is scored by a criterion that rewards the best ‘novice’ behavior.
There are many estimated kinks where AI powered test automation needs to be worked out. Here is a list of challenges and problems which you can face while implementing AI-powered test methodologies,
• It is difficult to identify every algorithm and perfecting them.
• You need extra ordinary amount of data to train a bot to perform test automation.
• You can only anticipate how your bot will behave from the high volume input data.
• Bots tend to repeat tasks even when the data feed are new.
• You will always need to train your bot and update their algorithms. It is a never ending process indeed.
There are also one or two AI techniques which can be used in software testing.
1) There is a Monte Carlo method. In this we look together at a simple example and plan to use it straight in web traffic.
2) We have a close look at the key features of the idea of Artificial Neural Network (structure, human brain inspiration, learning vs testing set, activation function) and talk over some current applications. Then we propose possible new applications like Graphical User Interface visual testing and memory/bugs usage/cross-module distribution check or traffic density.