For a machine-learning algorithm to be truly—and, in some cases, just marginally—effective, it needs to be trained using data sets that are as close to real-life collections as possible. However, when ...