runningCorrelation.py

#/usr/bin/env python
import math

class RunningCorrelation:
  def __init__(self):
    self.n = 0.0
    self.mean_x = 0.0
    self.mean_y = 0.0
    self.m2_x = 0.0
    self.m2_y = 0.0
    self.cov = 0.0
    self.variance_x = 0.0
    self.variance_y = 0.0
    self.old_mean_x = 0.0
    self.old_mean_y = 0.0
  
  def __online_variance(self,x,y):
    delta_x = x - self.mean_x
    delta_y = y - self.mean_y
  
    self.mean_x = self.mean_x + delta_x / self.n
    self.mean_y = self.mean_y + delta_y / self.n
    
    self.m2_x = self.m2_x + delta_x * (x - self.mean_x)  # This expression uses the new value of mean
    self.m2_y = self.m2_y + delta_y * (y - self.mean_y)  # This expression uses the new value of mean

    self.variance_x = self.m2_x/(self.n - 1)
    self.variance_y = self.m2_y/(self.n - 1)
  
  def __online_covariance(self,x,y):
    self.cov = self.cov + ((y - self.old_mean_y) * (x - self.old_mean_x)) * (self.n - 1.0) / self.n

  def push(self,x,y):
    self.n = self.n + 1
    
    if self.n == 1:
      self.mean_x = x
      self.mean_y = y
    else:
      self.old_mean_x = self.mean_x
      self.old_mean_y = self.mean_y
      
      self.__online_variance(x,y)
      self.__online_covariance(x,y)

  @property
  def r(self):
    if self.cov > 0:
     return self.cov / (math.sqrt(self.variance_x) * math.sqrt(self.variance_y))
    return 0
  
  @property
  def variance(self):
    return (self.variance_x,self.variance_y)

def main():
  rc = RunningCorrelation()
  v1 = [2.5,3.5,3.0,3.5,2.5,3.0]
  v2 = [3.0,3.5,1.5,5.0,3.5,3.0]
  
  for x,y in zip(v1,v2):
    rc.push(x,y)
    print rc.r #value of r is wrong, and is twice the correct value on the first pass > 1

if __name__ == '__main__':
  main()