Emgu 决策树

MCvDTreeParams

　　cvFolds　　　　　　　　//If this parameter is >1, the tree is pruned using cv_folds-fold cross validation.

　　maxCategories 　　　　//默认为10

　　maxDepth　　　　　　//This parameter specifies the maximum possible depth of the tree. That is the training algorithms attempts to split a node while its depth is less than max_depth. The actual depth may be smaller if the other termination criteria are met (see the outline of the training procedure in the beginning of the section), and/or if the tree is pruned.

　　minSampleCount　　　//A node is not split if the number of samples directed to the node is less than the parameter value.

　　priors　　　　　　　　//错分类代价

　　regressionAccuracy　　//Another stop criteria - only for regression trees. As soon as the estimated node value differs from the node training samples responses by less than the parameter value, the node is not split further.

　　truncatePrunedTree　　//If true, the cut off nodes (with Tn<=CvDTree::pruned_tree_idx) are physically removed from the tree. Otherwise they are kept, and by decreasing CvDTree::pruned_tree_idx (e.g. setting it to -1) it is still possible to get the results from the original un-pruned (or pruned less aggressively) tree.

　　use1seRule　　　　　　//If true, the tree is truncated a bit more by the pruning procedure. That leads to compact, and more resistant to the training data noise, but a bit less accurate decision tree.

　　useSurrogates　　　　//If true, surrogate splits are built. Surrogate splits are needed to handle missing measurements and for variable importance estimation.

public bool Train(

　　 Matrix<float> trainData,　　//A 32-bit floating-point, single-channel matrix,one vector per row

　　Emgu.CV.ML.MlEnum.DATA_LAYOUT_TYPE tflag,　　//data layout type:　　COL_SAMPLE or ROW_SAMPLE

　　Matrix<float> responses,　　//行阵等同于trainData，列数为1，表示训练数据的结果
　　Matrix<byte> varIdx,
　　　　//Can be null if not needed. When specified, identifies variables (features) of interest. It is a Matrix>int< of nx1

　　Matrix<byte> sampleIdx,　　　　//Can be null if not needed. When specified, identifies samples of interest. It is a Matrix>int< of nx1. 与上一个向量要么是基于0的整数列，要么是8位的标记，1预示有用，0预示跳过

　　Matrix<byte> varType,　　　　//The types of input variables. 是一个各个特征类型的基于0的标记（特征类型是CV_VAR_CATEGORICAL 还是 CV_VAR_ORDERED），它的长度是特征的个数加1，最后一个数字指定学习结果的类型

　　Matrix<byte> missingMask,　　//掩码矩阵，值为1表示相对应的值无效，值为0表示相对应的值有用

　　MCvDTreeParams param　　　　//The parameters for training the decision tree

)

示例程序：

　　private static DTree MushroomCreateDTree(Matrix<float> data, Matrix<byte> missing, Matrix<float> responses, int weight)

　　{

　　DTree dtree=new DTree();
　　Matrix<byte> varType=new Matrix<byte>(data.Width+1,1);
　　float[] priors = new float[] { 1, weight };
　　MCvDTreeParams mcd = new MCvDTreeParams();
　　mcd.maxDepth=8;
　　mcd.minSampleCount=10;
　　mcd.regressionAccuracy=0;
　　mcd.useSurrogates=true;
　　mcd.maxCategories=20;
　　mcd.cvFolds=10;
　　mcd.use1seRule=true;
　　mcd.truncatePrunedTree=true;
　　IntPtr p = Marshal.UnsafeAddrOfPinnedArrayElement(priors, 0);　//将float[] 转化成 IntPtr类型
　　mcd.priors=p;
　　dtree.Train(data,Emgu.CV.ML.MlEnum.DATA_LAYOUT_TYPE.ROW_SAMPLE,responses,null,null,varType,missing,mcd);

　　dtree.Save(System.Environment.CurrentDirectory + "\\dtee.xml");
　　int varSuccess = 0;
　　for (int i = 0; i < data.Rows; i++)
　　{
　　　　Matrix<float> sample=new Matrix<float>(1,data.Width);
　　　　Matrix<byte> missingDataMask=new Matrix<byte>(1,missing.Width);
　　　　for (int j = 0; j < data.Width; j++)
　　　　{
　　　　　　sample[0, j] = data[i, j];
　　　　}
　　　　for (int j = 0; j < missing.Width; j++)
　　　　{
　　　　　　missingDataMask[0, j] = missing[i, j];
　　　　}
　　　　MCvDTreeNode mcn = dtree.Predict(sample, missingDataMask, false);
　　　　if (mcn.value == responses[i, 0])
　　　　varSuccess++;
　　}
　　Console.WriteLine(responses.Rows +": " + varSuccess);

　　return dtree;

}