Source Code for Module mvpa.tests.test_stats

  1  # emacs: -*- mode: python; py-indent-offset: 4; indent-tabs-mode: nil -*- 
  2  # vi: set ft=python sts=4 ts=4 sw=4 et: 
  3  ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ## 
  4  # 
  5  #   See COPYING file distributed along with the PyMVPA package for the 
  6  #   copyright and license terms. 
  7  # 
  8  ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ## 
  9  """Unit tests for PyMVPA stats helpers""" 
 10   
 11  from mvpa.base import externals 
 12  from mvpa.clfs.stats import MCNullDist, FixedNullDist, NullDist 
 13  from mvpa.datasets import Dataset 
 14  from mvpa.measures.glm import GLM 
 15  from mvpa.measures.anova import OneWayAnova, CompoundOneWayAnova 
 16  from mvpa.misc.fx import doubleGammaHRF, singleGammaHRF 
 17  from tests_warehouse import * 
 18  from mvpa import cfg 
 19  from numpy.testing import assert_array_almost_equal 
 20   
 21  # Prepare few distributions to test 
 22  #kwargs = {'permutations':10, 'tail':'any'} 
 23  nulldist_sweep = [ MCNullDist(permutations=30, tail='any'), 
 24                     MCNullDist(permutations=30, tail='right')] 
 25   
 26  if externals.exists('scipy'): 
 27      from mvpa.support.stats import scipy 
 28      from scipy.stats import f_oneway 
 29      from mvpa.clfs.stats import rv_semifrozen 
 30      nulldist_sweep += [ MCNullDist(scipy.stats.norm, permutations=30, 
 31                                     tail='any'), 
 32                          MCNullDist(scipy.stats.norm, permutations=30, 
 33                                     tail='right'), 
 34                          MCNullDist(rv_semifrozen(scipy.stats.norm, loc=0), 
 35                                     permutations=30, tail='right'), 
 36                          MCNullDist(scipy.stats.expon, permutations=30, 
 37                                     tail='right'), 
 38                          FixedNullDist(scipy.stats.norm(0, 10.0), tail='any'), 
 39                          FixedNullDist(scipy.stats.norm(0, 10.0), tail='right'), 
 40                          scipy.stats.norm(0, 0.1) 
 41                          ] 
 42   
 43 -class StatsTests(unittest.TestCase): 
 44      """Unittests for various statistics""" 
 45   
 46   
 47      @sweepargs(null=nulldist_sweep[1:]) 
 48 -    def testNullDistProb(self, null): 
 49          """Testing null dist probability""" 
 50          if not isinstance(null, NullDist): 
 51              return 
 52          ds = datasets['uni2small'] 
 53   
 54          null.fit(OneWayAnova(), ds) 
 55   
 56          # check reasonable output. 
 57          # p-values for non-bogus features should significantly different, 
 58          # while bogus (0) not 
 59          prob = null.p([20, 0, 0, 0, 0, N.nan]) 
 60          # XXX this is labile! it also needs checking since the F-scores 
 61          # of the MCNullDists using normal distribution are apparently not 
 62          # distributed that way, hence the test often (if not always) fails. 
 63          if cfg.getboolean('tests', 'labile', default='yes'): 
 64              self.failUnless(N.abs(prob[0]) < 0.05, 
 65                              msg="Expected small p, got %g" % prob[0]) 
 66          if cfg.getboolean('tests', 'labile', default='yes'): 
 67              self.failUnless((N.abs(prob[1:]) > 0.05).all(), 
 68                              msg="Bogus features should have insignificant p." 
 69                              " Got %s" % (N.abs(prob[1:]),)) 
 70   
 71          # has to have matching shape 
 72          if not isinstance(null, FixedNullDist): 
 73              # Fixed dist is univariate ATM so it doesn't care 
 74              # about dimensionality and gives 1 output value 
 75              self.failUnlessRaises(ValueError, null.p, [5, 3, 4]) 
 76   
 77   
 78 -    def testAnova(self): 
 79          """Do some extended testing of OneWayAnova 
 80   
 81          in particular -- compound estimation 
 82          """ 
 83   
 84          m = OneWayAnova()               # default must be not compound ? 
 85          mc = CompoundOneWayAnova(combiner=None) 
 86          ds = datasets['uni2medium'] 
 87   
 88          # For 2 labels it must be identical for both and equal to 
 89          # simple OneWayAnova 
 90          a, ac = m(ds), mc(ds) 
 91   
 92          self.failUnless(a.shape == (ds.nfeatures,)) 
 93          self.failUnless(ac.shape == (ds.nfeatures, len(ds.uniquelabels))) 
 94   
 95          self.failUnless((ac[:, 0] == ac[:, 1]).all()) 
 96          self.failUnless((a == ac[:, 1]).all()) 
 97   
 98          ds = datasets['uni4large'] 
 99          ac = mc(ds) 
100   
101          if cfg.getboolean('tests', 'labile', default='yes'): 
102              # All non-bogus features must be high for a corresponding feature 
103              self.failUnless((ac[(N.array(ds.nonbogus_features), 
104                                   N.arange(4))] >= 1).all()) 
105          # All features should have slightly but different CompoundAnova 
106          # values. I really doubt that there will be a case when this 
107          # test would fail just to being 'labile' 
108          self.failUnless(N.max(N.std(ac, axis=1))>0, 
109                          msg='In compound anova, we should get different' 
110                          ' results for different labels. Got %s' % ac) 
111   
112 -def suite(): 
113      """Create the suite""" 
114      return unittest.makeSuite(StatsTests) 
115   
116   
117  if __name__ == '__main__': 
118      import runner 
119