After illustrating iterators based on folds, p-out, and custom schemes, we'll continue our overview on cross-validation iterators and quote all of the sampling-based ones.
The sampling schemes are different because they do not split the training set, but they sample it using different approaches: subsampling or bootstrapping.
Subsampling is performed when you randomly select a part of the available data, obtaining a smaller dataset than the initial one.
Subsampling is very useful, especially when you need to test your hypothesis extensively, but you prefer not to obtain your validation from extremely small test samples (so, you can opt out of a leave-one-out approach or a KFold using a large number of folds). The ...