Logic Regression - storage of trees


This file describes how trees are stored in files for further handling, as explained in the helpfile of the function logregtree() . When storing trees, we number the location of the nodes using the following scheme. This is an example for a tree with at most 8 terminal nodes, but the generalization should be obvious.





Each node may or may not be present in the current tree. If it is present, it can contain an operator (``and'' or ``or''), in which case it has to child nodes, or it can contain a variable, in which case the node is a terminal node. It is also possible that the node does not exist (as the user only specifies the maximum tree size, not the tree size that is actually fitted).
Output files have one line for each node. Each line contains 5 numbers:
  1. the number of the node
  2. does this node contain an ``and'' (1), an ``or'' (2), a variable (3), or is the node empty (0)
  3. if the node contains a variable, which one is it; e.g. if this number is 3 the node contains X3
  4. if the node contains a variable, does it contain the regular variable (0) or its complement (1)
  5. is the node empty (0) or not (1) (this information is redundant with the second number)

Example







The above tree is represented as

1 1 0 0 1
2 2 0 0 1
3 2 0 0 1
4 2 0 0 1
5 2 0 0 1
6 320 0 1
7 2 0 0 1
8 317 0 1
9 312 0 1
10 3 3 0 1
11 313 1 1
12 0 0 0 0
13 0 0 0 0
14 3 2 0 1
15 3 1 0 1




For questions please contact Ingo Ruczinski or Charles Kooperberg.