In Machine Learning, the training set error is often called the
*empirical error* or *empirical risk*, andn this is the error the
classifier incurs over the sample:

\begin{equation} L_S(h) = \frac{|\{i \in [m] : h(x_i) \ne y_i\}}{m} \end{equation}

Given a hypothesis class \(H\), finding the hypothesis \(h \in H\) that minimizes the empirical risk is a simple learning strategy.