I have recently begún to learn about model based recursive partitioning by playing around with MOB in the party package. I came across this mobForest package but am a little baffled towards what it is actually doing.
A MOB model (if I can call it that) is of the form
Y ~ X1,...,Xk | Z1,...,Zl using the MOB vignette's notation. In mobForest I would have a sequence of partitions (Z1,...,Zl ;...; A1,...Al) for which evey node would have a model of the form Y ~ X1,...Xk fitted to it. Is this an ensemble of MOB models or am I completely misunderstanding this?
Bootstraping the observations and sampling your variables at splits to build a sequence of trees helps tackle many problems that a single tree has. But how does that come into play here?
On a side note, when I use mobForestAnalysis to do model-based random forest analysis I dont know how to use this (Model?) to predict on new observations. Can someone help me out on this?