Two-stage learning

Typically, GMDH determines the coefficients from only a portion of the available dataset, while the rest gets used up for model performance evaluation. Our package additionaly offers a method that manages to calculate model coefficients using the entire dataset, in two stages.

The first stage is to determine optimal structure by some holdout method, using entire dataset. Which holdout method  will be used is determined by choice of dataProvider parameter of the learning algorithm (current choices are PercentageSplitHandler, CvHandler). In the second stage model coefficients get refitted, again using the entire dataset, while retaining the previously found structure of the model.

If the relearn parameter of the classifier is set to true, learning will take place in two stages. Otherwise, typical learning, which is generally faster, will follow. Note that CvHandler usage makes sense only with the two-stage learning; it's a waste of resources otherwise.