Machine learning is a popular field of computer science, which is also required in a modern industrial environment. Its capabilities make it possible to solve many types of problems with superhuman speed and accuracy, and with the libraries available today we can achieve these results with only a few lines of code. However on many fields the spreading of these solutions is hindered by the fact that programs are usually written in script languages (especially Python), and they are hard to integrate with existing business applications.
The ability to integrate these solutions would improve if the machine learning algorithm were available on a server, and the teaching and predicting functions were accessible with a well-defined API (Application Programming Interface). The aim of my thesis is to show by examples how an API like this can be made, and what steps lead from identifying a problem to not only solving it by machine learning, but integrating and using this solution on client-side.
The servers created by me present the usage of machine learning in three different areas, discussing the advantages and disadvantages of distributed architectures on this field. The created workflow and the presented tools can help in designing, creating and integrating one’s own solution to a problem. But these methods doesn’t only help with the given challenges, but can also be a guideline for the development process of any machine learning task.