A common way to approach a complex problem is to begin by reducing it down to a single performance metric. Choosing the right metric for each problem is a valuable skill. Blindly relying on a poorly chosen metric can cause no end of trouble. We will discuss various common metrics (such as accuracy, mean squared error, area under the ROC, log-loss, etc), the strengths and weaknesses of each, and how to know when to choose one of these metrics over another. [325]


Comments are closed.