Hybrid algorithm

A hybrid algorithm is an algorithm built by combining two or more other algorithms that each solve the same underlying problem. The combination can work in either of two ways: the hybrid may pick one of its component algorithms up front, based on some measured characteristic of the input data, or it may switch between the component algorithms partway through, as the run progresses and conditions change. The point of building a hybrid this way is almost always to combine the desired features of each component so that the resulting algorithm outperforms, or otherwise behaves better than, any single one of its parts taken alone; a common motivating case is combining an algorithm that is fast on typical inputs with one that has a better guaranteed worst case, switching to the safer algorithm only when the input looks likely to trigger the fast algorithm's weak case. It is important that hybrid algorithm is a narrower label than it might first sound: it does not cover every case of an algorithm being built out of simpler pieces, since a great many algorithms can be described as combinations of smaller sub-procedures. The term is reserved specifically for combining algorithms that solve the identical problem but differ from each other in some other respect, most notably in their performance characteristics.