True Risk

machine-learning

Definition

True Risk

True risk $R(h)$, $h$ being a model is the expected loss on all possible data, including future, unseen examples. In practice, $R(h)$ is incomputable because the true distribution $P(x,y)$ is unknown.