The ideal situation for block cross-validation where a large sample spans the predictive space, and structure in the observations can be used to tease out more-independent data folds. In practice, the independence achieved through blocking is often difficult to assess.

The ideal situation for block cross-validation where a large sample spans the predictive space, and structure in the observations can be used to tease out more-independent data folds. In practice, the independence achieved through blocking is often difficult to assess.