数据挖掘算法：灰度关联分析

算法实现简述

我使用C++的armadillo矩阵运算库实现了灰度关联分析，可以进行权重分析。

类的结构

/*灰度关联分析：判断一组向量与一个向量的关联程度
 核心思想：归一化、关联度越大，其曼哈顿距离越小
 */
 class GRA
 {
 public:
  GRA(double coeff_ = 0.5)
  {
   coeff = coeff_;
  }
  GRA& loadData(mat& data);
  GRA& loadData(mat& data, vec& mainData);
  GRA& setCoeff(double coeff_);
  GRA& caculate(std::string method = "firstNorm");
  rowvec& coeffVec();
 private:
  double coeff;//在[0,1]
  rowvec final_coeff;//最终的系数
  mat dat;
 };

类的实现

GRA & GRA::loadData(mat & data)
{
	dat = data;
	return *this;
}

GRA & GRA::loadData(mat & data, vec & mainData)
{
	dat.zeros(data.n_rows, data.n_cols + 1);
	dat.col(0) = mainData;
	dat.submat(0, 1, data.n_rows - 1, data.n_cols) = data;
	return *this;
}

GRA & GRA::setCoeff(double coeff_)
{
	coeff = coeff_;
	return *this;
}

GRA & GRA::caculate(std::string method)
{
	//初值归一化
	if (method == "firstNorm")
	{
		for (int i{ 1 }; i < dat.n_rows; ++i)
		{
			dat.row(i) = dat.row(i) / dat.row(0);
		}
	}
	//平均值归一化
	else if (method == "avgNorm")
	{
		dat = (dat - mean(dat)) / sqrt(cov(dat));
	}
	//不归一化
	else if (method == "none")
	{

	}
	//抛出异常
	else
	{
		throw Exception("GRA Caculate method error, check parameter \"none\" \"avgNorm\" \"firstNorm\".");
	}
	//求关联系数
	mat tmp{dat.n_rows,dat.n_cols};
	for (int i{ 0 }; i < dat.n_cols; ++i)
	{
		tmp.col(i) = dat.col(0) - dat.col(i);
	}
	tmp = abs(tmp);
#ifdef _DEBUG
	cout << dat << endl;
#endif
	tmp = (min(min(tmp)) + coeff * max(max(tmp))) / (tmp + coeff*max(max(tmp)));
#ifdef _DEBUG
	cout << tmp << endl;
#endif
	//求关联度
	final_coeff = mean(tmp);
	return *this;
}

rowvec & GRA::coeffVec()
{
	return final_coeff;
}