OpenCV码源笔记——Decision Tree决策树

来自OpenCV2.3.1 sample/c/mushroom.cpp

 

1.首先读入agaricus-lepiota.data的训练样本。

   样本中第一项是e或p代表有毒或无毒的标志位；其他是特征，可以把每个样本看做一个特征向量；

   cvSeqPush( seq, el_ptr );读入序列seq中，每一项都存储一个样本即特征向量；

   之后，把特征向量与标志位分别读入CvMat* data与CvMat* reponses中

   还有一个CvMat* missing保留丢失位当前小于0位置；

 

2.训练样本

    dtree = new CvDTree;
    dtree->train( data, CV_ROW_SAMPLE, responses, 0, 0, var_type, missing,
    CvDTreeParams( 8, // max depth
    10, // min sample count 样本数小于10时，停止分裂 
    0, // regression accuracy: N/A here；回归树的限制精度
    true, // compute surrogate split, as we have missing data；；为真时，计算missing data和变量的重要性
    15, // max number of categories (use sub-optimal algorithm for larger numbers)类型上限以保证计算速度。树会以次优分裂（suboptimal split）的形式生长。只对2种取值以上的树有意义
    10, // the number of cross-validation folds；If cv_folds > 1 then prune a tree with K-fold cross-validation where K is equal to cv_folds
    true, // use 1SE rule => smaller tree；If true 修剪树. 这将使树更紧凑,更能抵抗训练数据噪声,但有点不太准确
    true, // throw away the pruned tree branches
    priors //错分类的代价我们判断的：有毒VS无毒 错误的代价比 the array of priors, the bigger p_weight, the more attention
    // to the poisonous mushrooms
    // (a mushroom will be judjed to be poisonous with bigger chance)
    ));

 

3.

double r = dtree->predict( &sample, &mask )->value;//使用predict来预测样本，结果为 CvDTreeNode结构,dtree->predict(sample,mask)->value是分类情况下的类别或回归情况下的函数估计值；

4.interactive_classification通过人工输入特征来判断。

 

#include "opencv2/core/core_c.h"
#include "opencv2/ml/ml.hpp"
#include <stdio.h>

void help()
{
	printf("\nThis program demonstrated the use of OpenCV's decision tree function for learning and predicting data\n"
		"Usage :\n"
		"./mushroom <path to ag