Source Code for Module mvpa.mappers.zscore

 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  """Simple mapper to perform zscoring""" 
10   
11  __docformat__ = 'restructuredtext' 
12   
13  from mvpa.base import warning, externals 
14   
15  import numpy as N 
16  from mvpa.base.dochelpers import enhancedDocString 
17  from mvpa.mappers.base import ProjectionMapper 
18   
19  if externals.exists('scipy', raiseException=True): 
20      import scipy.sparse 
21      if externals.versions['scipy'] >= (0, 7, 0): 
22          _identity = scipy.sparse.identity 
23      else: 
24          _identity = scipy.sparse.spidentity 
25   
26   
27  __all__ = [ 'ZScoreMapper' ]            # just to don't leak all the evil ;) 
28   
29 -class ZScoreMapper(ProjectionMapper): 
30      """Mapper to project data into standardized values (z-scores). 
31   
32      After the mapper has been instantiated, it has to be train first. 
33   
34      Since it tries to reuse ProjectionMapper, invariant features will 
35      simply be assigned a std == 1, which would be equivalent to not 
36      standardizing them at all.  This is similar to not touching them 
37      at all, so similar to what zscore function currently does 
38      """ 
39 -    def __init__(self, 
40                   baselinelabels=None, 
41                   **kwargs): 
42          """Initialize ZScoreMapper 
43   
44          :Parameters: 
45            baselinelabels : None or list of int or string 
46              Used to compute mean and variance 
47              TODO: not in effect now and need to be adherent to all 
48              `ProjectionMapper`s 
49          """ 
50   
51          ProjectionMapper.__init__(self, **kwargs) 
52          if baselinelabels is not None: 
53              raise NotImplementedError, "Support for baseline labels " \ 
54                    "is not yet implemented in ZScoreMapper" 
55          self.baselinelabels = baselinelabels 
56          #self._var = None 
57   
58   
59      __doc__ = enhancedDocString('ZScoreMapper', locals(), ProjectionMapper) 
60   
61   
62 -    def _train(self, dataset): 
63          """Determine the diagonal matrix with coefficients for standartization 
64          """ 
65          samples = dataset.samples 
66          X = self._demeanData(samples) 
67          std = X.std(axis=0) 
68   
69          # ??? equivalent to not touching values at all, but we don't 
70          #     have such ability in ProjectionMapper atm afaik 
71          std[std == 0] = 1.0 
72          n = len(std) 
73   
74          # scipy or numpy manages to screw up: 
75          # or YOH is too tired?: 
76          # (Pydb) zsm._proj 
77          # <1x1 sparse matrix of type '<type 'numpy.float64'>' 
78          #         with 1 stored elements (space for 1) 
79          #         in Compressed Sparse Column format> 
80          # *(Pydb) (N.asmatrix(ds1.samples) - zsm._mean).shape 
81          # (120, 1) 
82          # *(Pydb) (N.asmatrix(ds1.samples) - zsm._mean) * zsm._proj 
83          # matrix([[-0.13047326]]) 
84          # 
85          # so we will handle case with n = 1 with regular non-sparse 
86          # matrices 
87          if n > 1: 
88              proj = _identity(n) 
89              proj.setdiag(1.0/std) 
90              recon = _identity(n) 
91              recon.setdiag(std) 
92              self._proj = proj 
93              self._recon = recon 
94          else: 
95              self._proj = N.matrix([[1.0/std[0]]]) 
96              self._recon = N.matrix([std]) 
97