Shulou(Shulou.com)06/02 Report--

Background

Since the birth of computer science, its main goal is computing (using computers to process a large amount of data) and automation (using computers instead of mechanical repetitive manual labor). Today, more than half a century later, we are surprised to find that these two categories are still guiding the development of computer science: big data and artificial intelligence (AI).

Automated testing is an application direction in the field of artificial intelligence, and like self-driving and robots, it is the application scenario of AI technology. From the perspective of the development of technology, automated testing has undergone four generations of development and changes. From the earliest idea of automated testing to the introduction of AI into automated testing, the core demand throughout its development is to effectively solve the problem of time and labor cost in automated testing.

1.png

For a long time in the past, automated testing has been limited to traditional test script drivers. Whether it is the Lua interface used by game developers, or the UI Test similar to Xcode, or the Python Script that simply manipulates Android devices through ADB/minicap, its essence is the automatic operation simulation of artificially defined rules. Although the traditional method has a large workload, it has a low threshold, easy to start, easy to debug and easy to modify. It has been the mainstream scheme of automated testing for a long time. The differences in the relevant mainstream solutions are usually limited to differences in the form of automated script interfaces or rule definitions.

However, there are some obvious problems with this manual definition of rules: frequent version iterations. Each version change often requires re-modification and adjustment of the script. It is difficult to support scenarios that are complex or random (such as game-to-game) through simple script adjustments. In fact, for highly random game products, there are usually not too many tests on the game-to-game process.

Part of the test script requires the product itself to provide the corresponding operation interface to obtain the internal data of the game. This kind of dedicated test interface is not usually provided in the official release version, so it is difficult to test with automated scripts for products that are really online, and can only be tested manually. This can be said to be the core reason why automated testing tools can not be commercialized and commercialized on a large scale.

AI and Automated testing

From the 1950s to the early 1970s, the research of artificial intelligence was in the "inference period". In 1981, Werbos proposed the multilayer perceptron and introduced the back propagation (BP) algorithm. Since then, the neural network has been developed rapidly. In 1986, the machine learning algorithm of decision tree was proposed by J.R.Quinlan, that is, ID3 algorithm. In 1995, Vapnik and Cortes proposed support vector machine (SVM), which has been popularized as a method that can make great progress in machine learning. In 2005, the three-tier NN model rose strongly, and many experts thoroughly activated deep learning in theory and practice. In 2012, the Hinton research group participated in the ImageNet image recognition competition for the first time and won the championship through the CNN network-AlexNet. The CNN network attracted the attention of many researchers. In 2015, the Deep residual Network (ResNet) was proposed and performed prominently in many competitions. In 2016, AIphaGo, developed by Google's DeepMind company, defeated the then go world champion Lee se-dol, becoming the first artificial intelligence robot to beat a human professional go player and the first go world champion, causing an uproar in world public opinion. More and more technical fields begin to try to develop with AI technology. Up to now, AI technology has mature technology and stable application in many aspects, such as image processing, speech recognition, art creation, autopilot and so on.

After Google DeepMind published several papers on game automation in 2015 and 2016, and AlphaGo had a precedent for defeating humans in board games, Tencent Interactive Entertainment Group's TuringLab team began to think about how to use the latest AI technology to solve the above problems and integrate them into automated testing tools. From 2017 to today, through the experiments of multiple products and the cooperation of many departments within Tencent, the TuringLab team has successfully developed a set of automated AI testing framework based on deep learning and applied it to the daily testing of many world-famous game products.

The core idea of AI technology is to use the existing historical data to train a better AI network, through the trained AI network to deal with the current input data. Unlike traditional programming methods, the network model trained by AI algorithm usually has better generalization. In the traditional automated testing, the maintenance cost of project testing is very high, as long as the version is updated or the function changes, it is possible to produce a lot of maintenance work. After the introduction of AI technology, its generalization can effectively reduce the maintenance cost of automated testing, programming in the AI model can not be changed, or directly increase the training data to complete the testing requirements.

In 2020, based on the deep learning of the automated AI testing framework, the TuringLab team sorted out its development and application experience into the book "AI Automated testing: technical principles, platform Construction and Engineering practice" to share with readers how Tencent Games combines AI technology with automated testing to complete the testing requirements.

Lead creative team

Tencent Interactive Entertainment Group TuringLab team

Led by returned technical experts from the United States, the group includes a number of PhDs in the fields of image recognition processing and machine learning, as well as experts specializing in engineering technology. At present, the AI SDK automated test platform developed by the laboratory has been successfully connected to dozens of Tencent's commercial games in operation, and serves WEST products, that is, Tongshou Q products, as well as a number of products from various game studios.

[editor in chief]

Zhang Like

Head of Tencent TuringLab Lab, senior AI system design expert; PhD in computer science from the University of Texas at San Antonio, has worked as a research and development engineer and project leader for Microsoft, BCG, Uber and other Silicon Valley startups.

[co-author]

Zhou Dajun: the head of the AI engineering team of Tencent TuringLab Lab, has more than 10 years of engineering and technical development experience. GAPS (stress testing Master), which is responsible for the development, won the first best tool award of Tencent, and applied for and participated in submitting more than 70 patents related to engineering, drones and AI.

Huang Chao, a senior AI algorithm researcher, has published more than 20 papers in important journals and conferences at home and abroad, and submitted 25 patents related to AI.

Li Xudong, a senior AI algorithm researcher, has published 15 academic papers and applied for 10 AI-related patents in important conferences and journals at home and abroad.

Shen Junfeng: senior software engineer with more than 10 years of engineering and technology development experience, has submitted 10 related technology patents

Wang Jiemei: a senior engineer of Tencent, has submitted 23 invention patents, including 2 international searches.

Yang Xia editor: Tencent senior engineer, has submitted 26 invention patents

A brief introduction to books

The book "AI Automated testing: technical principles, platform Building and Engineering practice" is divided into three parts.

The first part is the principle part, which focuses on the image recognition algorithm, reinforcement learning based on value function and strategy gradient, the research status of imitation learning and the application of AI algorithm in automatic testing. It also explains the problems related to Android equipment debugging in detail, which lays a foundation for the landing application of specific tools for follow-up learning.

2.png

The second part is the platform, which will explain the design and implementation of Tencent Games AI automation development tools for engineers with 117 pages of detailed introduction, including the docking with Android equipment, the data labeling process, the specific implementation of AI algorithm in game automation, the secondary development of AI SDK platform and so on.

The third part is the best practice chapter, which will show engineers how to use the test platform introduced in this book to complete all kinds of different requirements in the testing of actual game products through practice cases under different requirements scenarios. and you can try to achieve customization on this basis.

Conclusion

AI as a technical buzzword inscription in recent years, there are many enterprises and industries hit the "artificial intelligence" label manufacturing gimmicks to attract the attention of the public. Under the background of a variety of automated testing technologies, the automated testing method combined with AI technology has achieved the effect of improving production efficiency and saving labor costs in many appropriate scenarios.

From the current AI automated testing practice of Tencent, the cost savings of multi-resolution mobile phone tests such as compatibility, performance, regression and so on are the most obvious and effective. Now, the TuringLab team makes the AI test platform free and open source. If users feel that the existing algorithms or functions can not meet the testing needs, they can develop some functions and algorithms that are more suitable for their own business. The book "AI Automated testing: technical principles, platform Building and Engineering practice" will also better guide you to the development direction of AI+ automated testing. Click to buy now, and you can enjoy a 20% discount on limited time!

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.