Python Median Test

median.py

import numpy
import scipy
from scipy import stats

def median_test(samples1, samples2):
  median = numpy.median(samples1 + samples2)

  below1, above1 = count_partition(samples1, median)
  below2, above2 = count_partition(samples2, median)

  below_exp = (below1 + below2) / 2;
  above_exp = (above1 + above2) / 2;

  result = scipy.stats.chisquare(
    [below1, above1, below2, above2],
    f_exp=[below_exp, above_exp, below_exp, above_exp]
  );

  return {"diff": result[0], "confidence": (1 - result[0])}

def count_partition(samples, cut_value):
  smaller = 0.0;
  larger = 0.0;
  for v in samples:
    if v > cut_value:
      larger += 1.0
    else:
      smaller += 1.0
  smaller_scaled = normalize(smaller, samples)
  larger_scaled = normalize(larger, samples)
  return smaller_scaled, larger_scaled

def normalize(value, samples):
  return (value / len(samples)) + 1

#s1 = [20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 21, 21, 21, 21, 21]
#s2 = [20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 21, 21, 21, 21, 21, 21]

print "MORE 20s"
s1 = [20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 21, 21, 21, 21, 21, 21, 21, 21, 21]
s2 = [20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21]

r = median_test(s1, s2)
print r

print "MORE 21s"
s1 = [20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21]
s2 = [20, 20, 20, 20, 20, 20, 20, 20, 20, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21, 21]

r = median_test(s1, s2)
print r