Why does my mex-function for linear regression return r-squared values above 1?

Question

dave 2018-10-1

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/421703-why-does-my-mex-function-for-linear-regression-return-r-squared-values-above-1

评论： dave 2018-10-2

In a mex-file I have a function called ols() that is supposed to calculate a linear regression for two vectors, 'x' and 'y'. Unfortunately, every now and then ols() returns r-squared values that are higher than 1, so at some point there must be an error in my calculations...but I can't figure out where. Any help would be be greatly appreciated.

std::vector<float> ols(const std::vector<float>& x, const std::vector<float>& y){
    if(x.size() != y.size()){ 
        mexErrMsgTxt("Error: The length of vector 'x' does not equal the length of vector 'y'.");
    }   
      float sumx  = 0.0;                     
      float sumx2 = 0.0;                     
      float sumxy = 0.0;                    
      float sumy  = 0.0;                     
      float sumy2 = 0.0;                     
      float n = x.size();
      for (int i=0;i<n;i++){ 
          sumx  += x[i];       
          sumx2 += pow(x[i], 2);  
          sumxy += x[i] * y[i];
          sumy  += y[i];      
          sumy2 += pow(y[i], 2); 
      } 
      float denominator = (n * sumx2 - pow(sumx, 2));
      float slope     = (n * sumxy  -  sumx * sumy) / denominator;
      float intercept = (sumy * sumx2  -  sumx * sumxy) / denominator;
      float r         = (sumxy - sumx * sumy / n) /    
                          sqrt((sumx2 - pow(sumx, 2)/n) *
                          (sumy2 - pow(sumy, 2)/n));
      float rsquared  = pow(r, 2);
      std::vector<float> stats;
      stats.push_back(slope);
      stats.push_back(intercept);
      stats.push_back(rsquared);
      return stats;
  }

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

James Tursa 2018-10-2

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/421703-why-does-my-mex-function-for-linear-regression-return-r-squared-values-above-1#answer_339316

编辑：James Tursa 2018-10-2

I haven't run your code, but just looking at it you are using the worst algorithm numerically with single precision numbers, which is not a good combination. At the very least you should be using a better algorithm, and maybe doing the calculations in double precision. E.g., see this link:

https://en.wikipedia.org/wiki/Algorithms_for_calculating_variance