/// <summary>
/// 获取差别矩阵
/// </summary>
public DiffMatrix getDiffMatrix(List <double> valLength)
{
DiffMatrix matrix = new DiffMatrix(); //差别矩阵
int sampleNum = trainCollection.Count; //相似样本个数
//组合训练样本
for (int i = 0; i < sampleNum - 1; i++)
{
DataRow first = trainCollection[i];
for (int j = i + 1; j < sampleNum; j++)
{
DataRow second = trainCollection[j];
//计算差别
if (!first[0].ToString().Equals(second[0].ToString()))//决策属性不同
{
//产生差别矩阵元素
DiffMatrixElem elem = new DiffMatrixElem(i, j);
int keyNum = keyCollection.Count;
for (int k = 1; k < keyNum; k++)//第0号属性是决策属性，不是关键属性
{
//关键属性
DataColumn keyAttr = keyCollection[k];
if (keyAttr.DataType.ToString().Equals("System.Single") ||
keyAttr.DataType.ToString().Equals("System.Double") ||
keyAttr.DataType.ToString().Equals("System.Int32"))
{
//提取浮点数据
double a = Convert.ToDouble(first[k]);
double x = Convert.ToDouble(second[k]);
//计算差别程度
//double diff = Math.Abs(Math.Abs(x - a) / a);
double diff = 0;
if (valLength[k - 1] != 0)
{
diff = Math.Abs(x - a) / valLength[k - 1];
}
if (diff > 0) //有区别,没有约简，置为0
{
elem.addDiffAttr(k, diff); //添加差别属性
}
}
else
{
//文本二元化
if (!first[k].ToString().Equals(second[k].ToString()))
{
elem.addDiffAttr(k, 1);//添加差别属性
}
}
}
//添加差别元素
matrix.addElem(elem);
}
}
}
//matrix.write();
return(matrix);
}
/// <summary>
/// 主过程
/// </summary>
public void run()
{
List <DataRow> trainCollection = new List <DataRow>(); //训练样本
List <DataRow> testCollection = new List <DataRow>(); //测试样本
List <double> valLength = new List <double>(); //值空间
List <double> max = new List <double> (); //最大值
//提取测试样本和训练样本
DataColumnCollection attrCollection = classify(testCollection, trainCollection, valLength, max);
//构造筛选对象
filter = new Filter(trainCollection, attrCollection);
int testNum = testCollection.Count; //训练样本个数
DataRow testSample; //测试样本
double sucNum = 0; //测试成功的样本数
for (int i = 0; i < testNum; i++) //遍历测试样本
{
//根据相似样本集和关键属性构造约简对象
reduct = new Reduct(trainCollection, attrCollection);
List <double> dependance = reduct.calDependance(); //求出属性的依赖度
DiffMatrix matrix = reduct.getDiffMatrix(valLength); //获得差别矩阵
//获得专家参数
List <double> expertPara = new List <double>(); //专家参数集合
int attrNum = attrCollection.Count;
for (int k = 0; k < attrNum - 1; k++)
{
expertPara.Add(1);
}
//为关键属性加权
List <double> weights = matrix.weightedKeyAttr(dependance, expertPara);
//按照属性对属性排序
List <int> sortedIds = matrix.sortWeight(weights); //按照权值大小排序，记录关键属性顺序
//约简生成core集
List <int> core = matrix.reduce(sortedIds); //根据权值顺序约简差别矩阵，获得核心集合
if (core.Count == 0)
{
continue;
}
//测试样本
testSample = testCollection[i];
//求解针对测试样本相似度满足阈值的训练样本及其相似度
Dictionary <int, double> id_u = filter.getIdSimilarity(testSample, valLength, weights, max, core);
if (id_u.Count == 0)
{
continue;
}
//调用多类SVM进行决策
Strategy strategy = new Strategy(id_u, weights, core);
strategy.classify(trainCollection);
if (strategy.predict(testSample))
{
sucNum++;
}
}
Console.WriteLine("准确度=" + (int)(sucNum / testNum * 10000) / 100.0 + "%");
}
