Chapter 7. Preparing Data Pipelines for Predictive Analytics and Machine Learning
Advances in data processing technology have changed the way we think about pipelines and what you can accomplish in real time. These advances also apply to machine learning—in many cases, making a predictive analytics application real-time is a question of infrastructure. Although certain techniques are better suited to real-time analytics and tight training or scoring latency requirements, the challenges preventing adoption are largely related to infrastructure rather than machine learning theory. And even though many topics in machine learning are areas of active research, there are also many useful techniques that are already well understood and can immediately provide major business impacts when implemented correctly.
Figure 7-1 shows a machine learning pipeline applied to a real-time business problem. The top row of the diagram represents the operational component of the application; this is where the model is applied to automate real-time decision making. For instance, a user accesses a web page, and the application must choose a targeted advertisement to display in the time it takes the page to load. When applied to real-time business problems, the operational component of the application always has restrictive latency requirement.