Аннотация

It is interesting to watch trends in software development come and go, and to watch languages become fashionable, and then just as quickly fade away. As machine learning and AI began to reemerge a few years ago, it was easy to look upon the hype with a great deal of skepticism.
AlphaGo, a UK-based company, used deep learning to defeat the Go masters. Go is a Chinese board game that very complicated due to a huge number of combinations. Living in China at the time, there was a lot of discussion about the panicked Go masters who refused to play the machine for fear that their techniques would be exposed or "learned" by the machines.
An AI Poker Bot named Libratus individually defeated four top human professional players in 2017. This was surprising because poker is a difficult game for machines to master. In poker, unlike Go, there is a lot of unknown information, making it an "imperfect information" game.
Machine traders are replacing human traders at many of the large investment banks. The rise of the "quant" on Wall Street is well documented. Examining the job opportunities at investment banks reveals a trend favoring math majors, data scientists, and machine learning experts.
IBM’s Watson can do amazing things, such as fix the elevator before breaks, adjust the sprinkler system in the vineyard to optimize yield, and help oilfield workers manage a drilling rig.
Despite the hype, it was not until confronted with problems that were very difficult to solve with existing software tools that I began to explore and appreciate the power of machine learning techniques.
Today, after several years of gaining an understanding about what these new techniques can do, and how to apply them, I find myself thinking differently about each problem I encounter. Almost every piece of software can benefit in some way from machine learning techniques.
Developing machine learning software requires us to think differently about problems, resulting in a new way to partition our development efforts. However, change is good, and using machine learning with a data-driven development methodology can allow us to solve previously unsolvable problems.
In this book, I will describe what I have discovered along my journey. I hope that it can help you in your future software endeavors.