C++计算AUC

• 按预测值和标签升序排序
• 遍历数组，遇到负样本neg+1，遇到正样本sum+=neg
• 考虑到存在正负样本预测值相同，需要加0.5的情况，增加辅助变量tmp和val
• 理论看这里

#include <iostream>
#include <algorithm>
#include <vector>

using namespace std;

struct val_np {
  double val;
  char np;
};

bool auc_cmp(val_np a, val_np b) {
  if(a.val == b.val) return a.np < b.np;
  return a.val < b.val;
}

double AUC(vector<val_np> v) {
  sort(v.begin(), v.end(), auc_cmp);
  int neg = 0, tmp = 0, n = v.size(), m = 0;
  double val = 0, sum = 0;
  for(int i = 0; i < n; ++i) {
    if(v[i].np == 'p') ++m;
  }
  if(m == 0 || m == n) return 0;
  for(int i = 0; i < n; ++i) {
    // 大于之前的预测值，更新变量，如果是正样本tmp=0，是负样本tmp=1
    if(v[i].val != val) {
      val = v[i].val;
      neg += tmp;
      tmp = 1;
      // 正样本需要更新sum
      if(v[i].np == 'p') {
        tmp = 0;
        sum += neg;
      }
    } else {  
      // 等于之前的预测值且是正样本，sum加小于它的neg个和等于它的tmp个
      if(v[i].np == 'p') {
        sum += neg+0.5*tmp;
      } else {
        ++tmp;
      }
    }
  }
  return sum/m/(n-m);
}

int main(int argc, char **argv) {
	vector<val_np> v(8);
  v[0].val = 0.2, v[0].np = 'p';
  v[1].val = 0.2, v[1].np = 'n';
  v[2].val = 0.1, v[2].np = 'n';
  v[3].val = 0.2, v[3].np = 'n';
  v[4].val = 0.3, v[4].np = 'p';
  v[5].val = 0.2, v[5].np = 'p';
  v[6].val = 0.3, v[6].np = 'n';;
  v[7].val = 0.2, v[7].np = 'n';
  cout << AUC(v) << endl;

  return 0;
}