Machine Learning Compared to Traditional Software
Machine learning software is a special kind of software that finds patterns in data, learns from them, and even recreates these patterns on new data. Developing the machine learning software is, therefore, focused on finding the right data, matching it with the appropriate algorithm, and evaluating its performance. Traditional software, on the contrary, is developed with the algorithm in mind. Based on software requirements, programmers develop algorithms that solve specific tasks and then test them. Data is secondary, although not completely unimportant. Both types of software can co-exist in the same software system, but the programmer must ensure compatibility between them.
In this chapter, we’ll explore where these two types of software systems are most appropriate. We’ll learn about the software development processes that programmers use to create both types of software. We’ll also learn about the four classical types of machine learning software – rule-based learning, supervised learning, unsupervised learning, and reinforcement learning. Finally, we’ll learn about the different roles of data in traditional and machine learning software – as input to pre-programmed algorithms in traditional software and input to training models in machine learning software.
The best practices introduced in this chapter provide practical guidance on when to choose each type of software and how to assess the advantages and disadvantages of these types. By exploring a few modern examples, we’ll understand how to create an entire software system with machine learning algorithms at the center.
In this chapter, we’re going to cover the following main topics:
- Machine learning is not a traditional software
- Probability and software – how well do they go together?
- Testing and validation – the same but different
