import libsvm.svm;
import libsvm.svm_model;
import libsvm.svm_node;
import libsvm.svm_parameter;
import libsvm.svm_problem;

public class SVM {
 public static void main(String[] args) {
  // 定義訓(xùn)練集點(diǎn)a{10.0, 10.0} 和 點(diǎn)b{-10.0, -10.0}，對(duì)應(yīng)lable為{1.0, -1.0}
  List<Double> label = new ArrayList<Double>();
  List<svm_node[]> nodeSet = new ArrayList<svm_node[]>();
  getData(nodeSet, label, "file/train.txt");

  int dataRange=nodeSet.get(0).length;
  svm_node[][] datas = new svm_node[nodeSet.size()][dataRange]; // 訓(xùn)練集的向量表
  for (int i = 0; i < datas.length; i++) {
   for (int j = 0; j < dataRange; j++) {
    datas[i][j] = nodeSet.get(i)[j];
   }
  }
  double[] lables = new double[label.size()]; // a,b 對(duì)應(yīng)的lable
  for (int i = 0; i < lables.length; i++) {
   lables[i] = label.get(i);
  }

  // 定義svm_problem對(duì)象
  svm_problem problem = new svm_problem();
  problem.l = nodeSet.size(); // 向量個(gè)數(shù)
  problem.x = datas; // 訓(xùn)練集向量表
  problem.y = lables; // 對(duì)應(yīng)的lable數(shù)組

  // 定義svm_parameter對(duì)象
  svm_parameter param = new svm_parameter();
  param.svm_type = svm_parameter.EPSILON_SVR;
  param.kernel_type = svm_parameter.LINEAR;
  param.cache_size = 100;
  param.eps = 0.00001;
  param.C = 1.9;
  // 訓(xùn)練SVM分類模型
  System.out.println(svm.svm_check_parameter(problem, param));
  // 如果參數(shù)沒(méi)有問(wèn)題，則svm.svm_check_parameter()函數(shù)返回null,否則返回error描述。
  svm_model model = svm.svm_train(problem, param);
  // svm.svm_train()訓(xùn)練出SVM分類模型

  // 獲取測(cè)試數(shù)據(jù)
  List<Double> testlabel = new ArrayList<Double>();
  List<svm_node[]> testnodeSet = new ArrayList<svm_node[]>();
  getData(testnodeSet, testlabel, "file/test.txt");

  svm_node[][] testdatas = new svm_node[testnodeSet.size()][dataRange]; // 訓(xùn)練集的向量表
  for (int i = 0; i < testdatas.length; i++) {
   for (int j = 0; j < dataRange; j++) {
    testdatas[i][j] = testnodeSet.get(i)[j];
   }
  }
  double[] testlables = new double[testlabel.size()]; // a,b 對(duì)應(yīng)的lable
  for (int i = 0; i < testlables.length; i++) {
   testlables[i] = testlabel.get(i);
  }

  // 預(yù)測(cè)測(cè)試數(shù)據(jù)的lable
  double err = 0.0;
  for (int i = 0; i < testdatas.length; i++) {
   double truevalue = testlables[i];
   System.out.print(truevalue + " ");
   double predictValue = svm.svm_predict(model, testdatas[i]);
   System.out.println(predictValue);
   err += Math.abs(predictValue - truevalue);
  }
  System.out.println("err=" + err / datas.length);
 }

 public static void getData(List<svm_node[]> nodeSet, List<Double> label,
   String filename) {
  try {

   FileReader fr = new FileReader(new File(filename));
   BufferedReader br = new BufferedReader(fr);
   String line = null;
   while ((line = br.readLine()) != null) {
    String[] datas = line.split(",");
    svm_node[] vector = new svm_node[datas.length - 1];
    for (int i = 0; i < datas.length - 1; i++) {
     svm_node node = new svm_node();
     node.index = i + 1;
     node.value = Double.parseDouble(datas[i]);
     vector[i] = node;
    }
    nodeSet.add(vector);
    double lablevalue = Double.parseDouble(datas[datas.length - 1]);
    label.add(lablevalue);
   }
  } catch (Exception e) {
   e.printStackTrace();
  }

 }
}