Algorithms

Note

There are various methods for building redescriptions.

Siren allows to mine redescription using the greedy ReReMi algorithm as well as the splittrees and layeredtrees algorithms based on Classification and Regression Trees (CART).

For more details, check the original publications in the references section.

Greedy

ReReMi

The ReReMi algorithm [C.2] is a greedy update scheme for mining redescription that basically tries to find the best atomic extension at each step, constructing queries iteratively.

Tree-based

CARTWheels (a variant is available in Siren)

The first algorithm introduced for redescription mining was actually based on alternating between constructing CARTs and hence was called the CARTWheels algorithm [E.1].

See the little slideshow below to understand how redescriptions are constructed with this approach and refer to the original publication for more details.

Layeredtrees

An alternative method for constructing CARTs is to build them layer by layer. This method is implemented by the layeredtrees algorithm [D.2].

Splittrees

Finally the third method available in Siren construct queries by progressively increasing the depth of the trees. This method is implemented by the splittrees algorithm [D.2].