source: trunk/yat/utility/WeNNI.cc @ 1725

Last change on this file since 1725 was 1725, checked in by Jari Häkkinen, 12 years ago

Addresses #464. Weight zero will kill NaNs? and Infs.

  • Property svn:eol-style set to native
  • Property svn:keywords set to Author Date Id Revision
File size: 2.8 KB
Line 
1// $Id: WeNNI.cc 1725 2009-01-15 16:57:36Z jari $
2
3/*
4  Copyright (C) 2004 Jari Häkkinen
5  Copyright (C) 2005 Peter Johansson
6  Copyright (C) 2006 Jari Häkkinen
7  Copyright (C) 2007, 2008 Jari Häkkinen, Peter Johansson
8  Copyright (C) 2009 Jari Häkkinen
9
10  This file is part of the yat library, http://dev.thep.lu.se/yat
11
12  The yat library is free software; you can redistribute it and/or
13  modify it under the terms of the GNU General Public License as
14  published by the Free Software Foundation; either version 3 of the
15  License, or (at your option) any later version.
16
17  The yat library is distributed in the hope that it will be useful,
18  but WITHOUT ANY WARRANTY; without even the implied warranty of
19  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20  General Public License for more details.
21
22  You should have received a copy of the GNU General Public License
23  along with yat. If not, see <http://www.gnu.org/licenses/>.
24*/
25
26#include "WeNNI.h"
27#include "Matrix.h"
28#include "stl_utility.h"
29
30#include <algorithm>
31#include <cmath>
32#include <fstream>
33#include <limits>
34
35namespace theplu {
36namespace yat {
37namespace utility {
38
39
40  WeNNI::WeNNI(const utility::Matrix& matrix,const utility::Matrix& flag,
41               const unsigned int neighbours)
42    : NNI(matrix,flag,neighbours), imputed_data_raw_(matrix)
43  {
44    //estimate();
45  }
46
47
48  // \hat{x_{ij}}=\frac{ \sum_{k=1,N} \frac{w_{kj}*x_{kj}}{d_{ki}} }
49  //                   { \sum_{k=1,N} \frac{w_{kj}       }{d_{ki}} }
50  // where N is defined in the paper cited in the NNI class definition
51  // documentation.
52  unsigned int WeNNI::estimate(void)
53  {
54    double small_number=std::numeric_limits<double>::epsilon();
55    for (size_t i=0; i<data_.rows(); i++) {
56      std::vector<std::pair<size_t,double> > distance(calculate_distances(i));
57      std::sort(distance.begin(),distance.end(),
58                pair_value_compare<size_t,double>());
59      bool row_imputed=true;
60      for (size_t j=0; j<data_.columns(); j++) {
61        std::vector<size_t> knn=nearest_neighbours(j,distance);
62        double new_value=0.0;
63        double norm=0.0;
64        for (std::vector<size_t>::const_iterator k=knn.begin(); k!=knn.end();
65             ++k) {
66          // Avoid division with zero (perfect match vectors)
67          double d=(distance[*k].second ? distance[*k].second : small_number);
68          double w=weight_(distance[*k].first,j)/d;
69          if (w) {
70            new_value += w*data_(distance[*k].first,j);
71            norm      += w;
72          }
73        }
74        // No impute if no contributions from neighbours.
75        if (norm) {
76          imputed_data_raw_(i,j) = new_value/norm;
77          double w=weight_(i,j);
78          if (w)
79            imputed_data_(i,j) = w*data_(i,j) + (1-w)*imputed_data_raw_(i,j);
80          else
81            imputed_data_(i,j) = imputed_data_raw_(i,j);
82        }
83        else
84          row_imputed=false;
85      }
86      if (!row_imputed)
87        not_imputed_.push_back(i);
88    }
89    return not_imputed_.size();
90  }
91
92
93}}} // of namespace utility, yat, and theplu
Note: See TracBrowser for help on using the repository browser.