Point Predictions

A point prediction for a label $y\in\mathbf{Y}$ (and element of a label space $\mathbf{Y}$) is an element $\hat y\in\mathbf{Y}$. The quality of the point prediction $\hat y$ is measured by a distance between $\hat y$ and the actual label $y$. Other kinds of predictions used in conformal prediction are prediction sets and predictive distributions. In competitive on-line prediction, much more general predictions are used: see game of prediction.