Source code for ClearMap.Analysis.Statistics.StatisticalTests

# -*- coding: utf-8 -*-
"""
statisticalTests
================

Colelction of some statistics tests
"""
__author__    = 'Christoph Kirst <christoph.kirst.ck@gmail.com>'
__license__   = 'GPLv3 - GNU General Pulic License v3 (see LICENSE.txt)'
__copyright__ = 'Copyright © 2020 by Christoph Kirst'
__webpage__   = 'http://idisco.info'
__download__  = 'http://www.github.com/ChristophKirst/ClearMap2'

import sys
import numpy as np

from scipy.stats import rankdata
from scipy.stats import distributions



[docs]
def test_cramer_von_mises_2_sample(x, y):
  """
  Computes the Cramer von Mises two sample test.

  Arguments
  ---------
  x, y : 1-D ndarrays
    Two arrays of sample observations assumed to be drawn from a continuous 
    distribution, sample sizes can be different.

  Returns
  -------
  t_statistic : float
    The t-statistic.
  p_value : float 
    Two-tailed p-value.
      
  Notes
  -----
  This is a two-sided test for the null hypothesis that 2 independent samples
  are drawn from the same continuous distribution.    
  
  References
  ----------
  * modified from https://github.com/scipy/scipy/pull/3659
  """
  
  #following notation of Anderson et al. doi:10.1214/aoms/1177704477
  N = len(x)
  M = len(y)
  assert N * M * (N + M) < sys.float_info.max
  
  alldata = np.concatenate((x,y))
  allranks = rankdata(alldata)
  ri = allranks[:N]
  sj = allranks[-M:]
  
  i = rankdata(x)
  j = rankdata(y)
  #Anderson et al. Eqn 10
  U = N*np.sum((ri - i)**2) + M*np.sum((sj - j)**2)
  #print U
  
  #Anderson et al. Eqn 9
  T = U/(N * M * (N + M)) - (4 * M * N - 1)/(6 * (M + N))
  #print T
  
  Texpected = 1./6 + 1./(6 * (M + N))
  Tvariance = 1./45 * (M + N + 1)/(M + N)**2 * (4 * M * N * (M+N) - 3*(M**2 + N**2) - 2*M*N)/(4 * M * N)
  
  zscore = np.abs(T - Texpected) / np.sqrt(Tvariance)
  #print zscore
  
  return T, 2*distributions.norm.sf(zscore)