PMCRI uses a variant of the Prism algorithm described in Chapter 11,
called PrismTCS, but the differences are not important here. The important
point is how the CDM model is used to control the rule generation process.
Assuming that there are p processors, all roughly identical, the sample selection
procedure at Layer 1 divides the data up approximately evenly amongst them.
If we focus on landscape-style data, that is achieved by giving each processor
all the instances for 1/p th of the total number of attributes.