Adaboost.cpp

#include "Adaboost.h"
#define MAX 1000000
#define MIN -100000

namespace MLL
{
	double* Adaboost::rangeSize(const int &axis)
	{
		double rangemin = MAX;
		double rangemax = MIN;
		double *range = (double *)malloc(sizeof(double) * 2);
		for(size_t i = 0; i < data.size(); i++)
		{
			if(data[i][axis] > rangemax)
				rangemax = data[i][axis];
			if(data[i][axis] < rangemin)
				rangemin = data[i][axis];
		}
		range[0] = rangemin;
		range[1] = rangemax;
		return range;
	}
	std::vector<int> Adaboost::Classify(const int &axis, const double &threshVal, const std::string &threshIneq)
	{
		std::vector<int> label;
		for(size_t i = 0; i < data.size(); i++)
		{
			label.push_back(1);
			if(threshIneq == "lt")
			{
				if(data[i][axis] <= threshVal)
					label[i] = -1;
			}
			else
			{
				if(data[i][axis] > threshVal)
					label[i] = -1;
			}
		}
		return label;
	}
	Adaboost::Stump Adaboost::buildStump(std::vector<double> &weight)
	{
		Stump stump;
		int i = 0, j = 0, k = 0, l = 0;
		double *errArr = (double *)malloc(sizeof(double)*data.size());
		double minErr = MAX;
		double weightError = 0;
		double *range = (double *)malloc(sizeof(double)*2);
		double rangemin = 0, rangemax = 0;
		double stepSize = 0;
		int numSteps = 2;//划分步长
		//int numSteps = data.size()/10;
		double threshVal = 0 ;
		int label_index = data[0].size()-1;
		std::vector<int> label;
		std::string threshIneq[2]= {"lt","gt"};
		for(i = 0; i < data[0].size()-1; i++) //属性
		{
			range = rangeSize(i);
			rangemin = range[0];
			rangemax = range[1];
			stepSize = (rangemax-rangemin) / numSteps;
			for(j = 0; j <= numSteps; j++) //其实是numstep+1步
			{
				threshVal = rangemin+stepSize*j;
				for(k = 0; k < 2; k++) //大于小于
				{
					label = Classify(i, threshVal, threshIneq[k]);
					weightError = 0;
					for(l = 0; l < data.size(); l++)
					{
						errArr[l] = label[l] - data[l][label_index] > 0 ? (label[l] - data[l][label_index]) / 2 : - (label[l]-data[l][label_index]) / 2;
						weightError += errArr[l] * weight[l];
					}
					if(weightError < minErr)
					{
						minErr = weightError;
						for(l = 0; l < data.size(); l++)
						{
							if(label[l] > 0)
								stump.twosubdata.left.push_back(data[l]);
							else
								stump.twosubdata.right.push_back(data[l]);
						}
						stump.label = label;
						stump.bestIndex = i;
						stump.minErr = minErr;
						stump.threshVal = threshVal;
						stump.ltOrgt = threshIneq[k];
						stump.alpha = 0.5 * log((1 - minErr) / minErr);
					}
				}
			}
		}
		return stump;
	}
	int Adaboost::AdaboostTrain()
	{
		int i = 0, j = 0, k = 0;
		int label_index = data[0].size()-1;
		std::vector<double> weight;
		double weightSum = 0;
		std::vector<double> aggErr;
		double sumErr = 0;
		for(i = 0; i < data.size(); i++)
		{
			weight.push_back(1.0 / data.size());
			aggErr.push_back(0);
		}
		for(i = 0; i < numIt; i++)
		{
			stump[i] = buildStump(weight);//写成赋值语句结构体中的结构体复制出现问题
			std::cout << "alpha====" << stump[i].alpha << std::endl;
			std::cout << "minErr=" << stump[i].minErr << std::endl;
			weightSum = 0;
			for(j = 0; j < data.size(); j++)
			{
				weight[j] = weight[j] * (exp(-1 * stump[i].alpha * (data[i][label_index] * stump[i].label[j])));
				weightSum += weight[j];
			}
			for(j = 0; j < data.size(); j++)
			{
				weight[j] /= weightSum;
			}
			for(j = 0; j < data.size(); j++)
			{
				for(k = 0; k < i; k++)
				{
					aggErr[j] += stump[k].alpha * (stump[k].label[j]);
				}
				if(aggErr[j] > 0)
					aggErr[j] = 1;
				else
					aggErr[j] = -1;
			}
			sumErr = 0;
			for(j = 0; j<data.size(); j++)
			{
				sumErr += (aggErr[j] - data[j][label_index]) >0 ? (aggErr[j]-data[j][label_index])/2 : -(aggErr[j]-data[j][label_index])/2;
			}
			//if(sumErr/data.size()<0.01)
			{
				std::cout<< "sumErr:" << sumErr<<std::endl;
				//break;
			}
		}
		for(i = 0; i < numIt; i++)
		{
			std::cout << "weight:" << stump[i].alpha << "\t" << "bestindex:" << stump[i].bestIndex << "\t" << "threshval:" << stump[i].threshVal << "\t" << "ltOrGt:" << stump[i].ltOrgt << "\t" << "error:" << stump[i].minErr <<std::endl;
		}
		return 0;
	}
	Adaboost::Adaboost(const std::string &file,const int &numIt)
	{
		this->numIt = numIt;
		stump.resize(numIt);
		LoadData(data,file);
		int label_index=data[0].size()-1;
		for(int l=0; l<data.size(); l++)
		{
			if(data[l][label_index] == 0)
			{
				data[l][label_index] = -1;
			}
		}
	}
}