如何在分类决策树中获取特定类类型的决策路径

时间:2015-10-01 11:30:26

标签: matlab classification instantiation decision-tree data-generation

假设我创建了一个分类决策树,如下所示:

HP(1:size(HP), end) = 0; LP(1:size(LP), end) = 1;
% the dt's input & target pop
x = [HP(:,1:end-1); LP(:,1:end-1)]; t = [HP(:,end); LP(:,end)];
dt = fitctree(x,t);
view(dt)
view(dt, 'mode', 'graph');

输出如下:

1  if x2<-21.4866 then node 2 elseif x2>=-21.4866 then node 3 else 1
2  class = 1
3  if x2<20.093 then node 4 elseif x2>=20.093 then node 5 else 0
4  if x1<27.8438 then node 6 elseif x1>=27.8438 then node 7 else 0
5  if x2<39.6866 then node 8 elseif x2>=39.6866 then node 9 else 1
6  if x1<-33.7504 then node 10 elseif x1>=-33.7504 then node 11 else 0
7  class = 1
8  class = 1
9  class = 0
10  class = 1
11  if x2<1.53772 then node 12 elseif x2>=1.53772 then node 13 else 0
12  if x2<-2.50063 then node 14 elseif x2>=-2.50063 then node 15 else 0
13  class = 0
14  if x1<14.2153 then node 16 elseif x1>=14.2153 then node 17 else 0
15  class = 1
16  class = 0
17  class = 1

并且

enter image description here

问题

1)如何获得导致值为“0”的叶子的所有路径?

2)是否存在基于决策树创建新实例的精细方式(除了随机生成实例和循环直到所需输出)?例如,我想创建一个随机实例,上面的树将其分类为'0'

1 个答案:

答案 0 :(得分:0)

如果您知道决策的结束节点,则可以找到路径。您可以从[lable, score, node, cusum] = predict(mdl,x);获取结束节点号。如果您想从Treebagger获取每棵树的节点,则需要为每棵树循环运行相同的节点。 [lable, score, node, cusum] = mdl.Trees{i}.predict(x);