Preventing Combinatorial Explosion

Selectional Combinatorial GMDH as described above is not offered as a Classifier in Weka by default. A version where only the nodes from the topmost layer and first two layers (layer of attributes included) can be used to build more complex models is offered. This modification makes further growth of existing network feasible regardless of its current number of layers, and, according to [1], the convergence properties of the algorithm are retained.