I am having trouble understanding Random Forest, especially some terms.
What is a node what is node size? What are predicted variables and split variables? Could someone clarify or point me somewhere where this is explained?
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
8
A forest is made out of trees, a tree is a graph and the points connected in Graphs are called nodes. See https://en.wikipedia.org/wiki/Graph_(abstract_data_type)
So in a decision tree or regression tree, each check of a predictor variable is a node. A split variable is the variable whose value decides, which branch to follow next. It is what you'd call an Independent variable in other regression techniques. Predictes variable is what the name says: The dependend variable.
-
$\begingroup$ Okay. So far understood. Why I wonder is how can a variable whose value decides, which branch to follow next. (that could be GINI i.e. i suppose?). $\endgroup$ Commented Jun 30, 2016 at 13:21
-
$\begingroup$ Sorry, could you rephrase the question, possible add a verb? "How can a variable..." do what? $\endgroup$– BernhardCommented Jun 30, 2016 at 13:23
-
$\begingroup$ Sorry. How Can a variable decide which branch to follow? that means, there are many branches, and in the end, I only follow one branch? I thought RF creates many many trees and in the end each trees casts a vote what class it could be? How can you follow a branch there? $\endgroup$ Commented Jun 30, 2016 at 13:25
-
$\begingroup$ In every single tree there are a lot of branches. You follow from the stem to the leafs of a tree by making a lot of decisions. Every time any branch in any tree is split, there is a decision. Each of these decisions is based on the value of a split variable. There is only a finite number of variables on which splits of branches can be made in the whole forest: The predictor variables. $\endgroup$– BernhardCommented Jun 30, 2016 at 13:32
-
$\begingroup$ Thanks for sticking with me. Its still not clear to me what predictor variables are. $\endgroup$ Commented Jun 30, 2016 at 13:37