学习笔记，仅供参考，有错必纠
回归树与基于规则模型
基于规则的模型
A rule(规则) is defined as a distinct path through a tree(树中一条不重复的路径).For the
tree, a new sample can only travel down a single path through the tree defined by these rules.(对于一棵树，新观测只能沿着唯一的一条路径由上至下，而路径正是由一系列规则定义的) The number of samples affected by a rule is called its coverage.(被一条规则影响的观测数，被称为它的覆盖)
Holmes等人构建了一种从模型树中建立规则的方法，他们使用了"分而治之"的策略
This procedure derives rules from many different model trees instead of from a single tree.
First, an initial model tree is created (they recommend using unsmoothed model trees). However, only the rule with the largest coverage is saved from this model. The samples covered by the rule are removed from the training set and another model tree is created with the remaining data. Again, only the rule with the maximum coverage is retained. This process repeats until all the training set data have been covered by at least one rule(被最后一条规则覆盖). A new sample is predicted by determining which rule(s) it falls under then applies the linear model associated with the largest coverage(要预测新样本时，首先将决定样本落入了哪些规则之中，然后使用其中覆盖最大的那条规则对应的线性模型进行预测).
