07-30-2013 12:01 PM
I am new to enterprise miner but I am finding it very useful for my work.
I am currently building a decision tree and am building them autonomously and by hand to compare. When building by hand I select my first two splits
as I want certain variables to be at the top. I then automatically complete them so I have the maximal tree. I then view the subtree plots and tables which
tells me the optimal size as having 13 leaf nodes.
How do I then view this optimal tree and take it forward? I can run autonomously and choose the option of creating the 13 leafed tree but that's not what I want as this wouldn't have the two variables I selected first to be at the top. I can right click on the diagram in the editor at each decision point and prune but this just deletes everything below it and I have no idea which 13 leaves I'm supposed to keep and I have too may variables to manually build every 13 leafed node. I am using version 12.1 and I could be missing the obvious but I can't seem to figure it out.
Thanks in advance
07-30-2013 01:09 PM
There are two ways to build decision trees.
1/ automatic it will build the autonously with no manual interaction.
2/ interactive (choose it by the ellisis at het menu/properies).
By this you can build it manually and then proceed (make active) at a leaflet to have it build to lower levels.
Do not forget to save before returning as the interactive mode is partially running on your desktop. (This is an exception in Miner)
07-31-2013 04:30 AM
Thanks for the reply. I know how to build the decision trees using both those methods.
My question involved once you have a maximal tree when building manually how do you select the optimal tree. For example my maximal tree has around 40 nodes. When I investigate the subtree plots/tables it shows me that the tree with 13 nodes is the optimum model. However I want to be able to see that tree. How do I select the 13 leafed tree to be used?
07-31-2013 04:46 AM
That is te pruning /back and stopping citeria. And than rerun make/ a copy.
That is something not quite intuitive (forgot the details). Not able at the moment to do a recap/replay.