Online Machine Learning: A Practical Guide with Examples in Python

Why OML? Among other things, it is about the decisive time advantage. This can be months, weeks, days, hours, or even just seconds. This time advantage can arise if Artificial Intelligence (AI) can evaluate data continuously, i.e., online. It does not have to wait until a complete set of data is available, but can already use a single observation to update the model. Does OML have other advantages besides the obvious time advantage? If so, what are they? We ask, are there limitations of BML that OML overcomes? It needs to be carefully examined at what price one gets these advantages from OML. How high is the memory requirement compared to conventional methods? Memory requirements also mean financial costs, e.g., due to higher energy requirements. Is OML possibly energy-saving and thus more sustainable, i.e., Green IT? Is it possible to obtain comparably good results? Does the quality (performance) suffer, do the results become less accurate? In order to answer these questions reliably, we first give an understandable introduction to OML in the theoretical part, which is suitable for beginners as well as for advanced users. Then we justify the criteria we found for the comparability of OML and BML, namely a well-comprehensible representation of quality, time, and memory requirements. In the second part, we address the question of exactly how OML can be used in practice. We are joined by experts from the field who report on their practical experiences, e.g., requirements for official statistics. We give