What predictor variables are used for a given pruning level in a classregtree?

I'm working with classification and regression trees - I can define them, prune them, and even view the pruned tree with the predictor variable names displayed using the view() function. However, for an optimally pruned tree, I want to know which of the 6 predictor variables are actually being used. I can see this in the view(Tree) display, but I need some sort of index to be returned in my code, so I don't have to manually open a tree, assess the variables, and store which ones ended up being significant every time. Is there a way to do this?

To reiterate - for a classregtree and a given pruning level, how can I extract the predictor variables (either a name or index) that are being used? My optimally pruned trees tend to disregard at least one of my 6 predictor variables, and I want to know which are actually used without manually viewing the tree.

I'm using Matlab 7.12.0.635 win64 Thank you very much!

Answer 1

[Ilya]

Prune the tree to the desired level using the prune method. Then use the cutvar method to see what variables are used for splits in each node. For example:

load ionosphere
t = classregtree(X,Y);
t = prune(t,'level',3);
varsUsedInTree = ismember(names(t),cutvar(t)')

Typing methods(t) or looking at the classregtree doc would give you a full list of methods.

[Rebecca]

Thank you for your help! I did not notice the cutvar method before. We ended up using:

t = classregtree(X, Y,'names',cellArray);
[~,~,~,bestLevel] = test(t, 'crossvalidate', X, Y);
t = prune(t, 'level', bestLevel);
varsUsedInTree = t.names(ismember(t.names,cutvar(Tree2)));

[cev]

when I run your code I get this message in Matlab 7.10.0(R2010a)

??? Undefined function or method 'names' for input arguments of type 'classregtree'.

Error in ==> Untitled2 at 4 varsUsedInTree = ismember(names(t),cutvar(t)')

[Ilya]

Replace

ismember(names(t),cutvar(t)')

with

ismember(t.names,cutvar(t))