source: prextra/dectreec.m @ 50

%
%          W = DECTREEC(A,N)
%
% Train a decision tree on A, using random feature subsets of size N.
% When N=0, no feature subsets are used.
% The decision tree training grows a full tree (no pruning), by
% splitting a single feature using one threshold. For the splitting the
% optimal GINI index is used.
function w = dectreec(a,featsubset)

if nargin<2
        featsubset = 0;
end
if nargin<1 || isempty(a)
        w = mapping(mfilename,{featsubset});
        w = setname(w,'Decision tree');
        return
end

if ~ismapping(featsubset)
        y = getnlab(a);
        opt.K = max(y);
        opt.featsubset = featsubset;
        if exist('decisiontree')==3
                v = decisiontree(+a,y,opt.K,featsubset);
        else
                v = tree_train(+a,y,opt);
        end
        w = mapping(mfilename,'trained',v,getlablist(a),size(a,2),opt.K);
else
        v = getdata(featsubset);
        n = size(a,1);
        if exist('decisiontree')==3

                if ~isa(v,'double')
                        error('This tree should have been trained with the C-code');
                end
                out = decisiontree(v,+a);
        else
                if ~isa(v,'cell')
                        error('This tree should have been trained with the Matlab code');
                end
                out = tree_eval(v,+a);
        end
        out = accumarray([(1:n)' out],ones(n,1));

        w = setdat(a,out,featsubset);
end
return