source: trunk/test/normalization_test.cc @ 2146

Last change on this file since 2146 was 2146, checked in by Peter, 12 years ago

fixes #580

  • Property svn:eol-style set to native
  • Property svn:keywords set to Id
File size: 12.9 KB
Line 
1// $Id: normalization_test.cc 2146 2010-01-15 23:32:09Z peter $
2
3/*
4  Copyright (C) 2008, 2009 Jari Häkkinen, Peter Johansson
5
6  This file is part of the yat library, http://dev.thep.lu.se/yat
7
8  The yat library is free software; you can redistribute it and/or
9  modify it under the terms of the GNU General Public License as
10  published by the Free Software Foundation; either version 3 of the
11  License, or (at your option) any later version.
12
13  The yat library is distributed in the hope that it will be useful,
14  but WITHOUT ANY WARRANTY; without even the implied warranty of
15  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16  General Public License for more details.
17
18  You should have received a copy of the GNU General Public License
19  along with yat. If not, see <http://www.gnu.org/licenses/>.
20*/
21
22#include "Suite.h"
23
24#include "yat/normalizer/Centralizer.h"
25#include "yat/normalizer/ColumnNormalizer.h"
26#include "yat/normalizer/Gauss.h"
27#include "yat/normalizer/qQuantileNormalizer.h"
28#include "yat/normalizer/QuantileNormalizer.h"
29#include "yat/normalizer/RowNormalizer.h"
30#include "yat/normalizer/Spearman.h"
31#include "yat/normalizer/Zscore.h"
32
33#include "yat/utility/DataIterator.h"
34#include "yat/utility/FileUtil.h"
35#include "yat/utility/Matrix.h"
36#include "yat/utility/MatrixWeighted.h"
37#include "yat/utility/WeightIterator.h"
38
39#include <boost/concept_archetype.hpp>
40
41#include <climits>
42#include <fstream>
43#include <limits>
44#include <vector>
45
46using namespace theplu::yat;
47void test_centralizer(test::Suite&);
48void test_column_normalize(test::Suite&);
49void test_gauss_normalize(test::Suite&);
50void test_qquantile_normalize(test::Suite&);
51void test_qquantile_normalize_weighted(test::Suite&);
52void test_quantile_normalize(test::Suite&);
53void test_row_normalize(test::Suite&);
54void test_spearman(test::Suite&);
55void test_spearman_weighted(test::Suite&);
56void test_z_score(test::Suite&);
57
58int main(int argc, char* argv[])
59{ 
60  test::Suite suite(argc, argv);
61  suite.err() << "testing normalizations ... " << std::endl;
62
63  test_centralizer(suite);
64  test_column_normalize(suite);
65  test_qquantile_normalize(suite);
66  test_qquantile_normalize_weighted(suite);
67  test_quantile_normalize(suite);
68  test_gauss_normalize(suite);
69  test_row_normalize(suite);
70  test_spearman(suite);
71  test_z_score(suite);
72
73  return suite.return_value();
74}
75
76
77void test_centralizer(test::Suite& suite)
78{
79  suite.err() << "Testing Centralizer\n";
80  std::vector<double> vec;
81  vec.push_back(1);
82  vec.push_back(2);
83  vec.push_back(3);
84  normalizer::Centralizer<> c;
85  c(vec.begin(), vec.end(), vec.begin());
86  for (size_t i=0; i<vec.size(); ++i)
87    suite.add(suite.equal(vec[i], static_cast<double>(i)-1.0));
88
89  std::vector<utility::DataWeight> vec2;
90  vec2.push_back(utility::DataWeight(1,1));
91  vec2.push_back(utility::DataWeight(2,0.5));
92  vec2.push_back(utility::DataWeight(2,0.5));
93  std::vector<utility::DataWeight> vec3(vec2.size());
94  c(vec2.begin(), vec2.end(), vec3.begin());
95  for (size_t i=0; i<vec2.size(); ++i)
96    suite.add(suite.equal(vec3[i].weight(), vec2[i].weight()));
97  suite.add(suite.equal(vec3[0].data(), -0.5));
98  suite.add(suite.equal(vec3[1].data(), 0.5));
99  suite.add(suite.equal(vec3[2].data(), 0.5));
100
101  // compile test should not be run
102  if (false) {
103    boost::detail::dummy_constructor dummy_cons;
104    c(boost::input_iterator_archetype<double>(), 
105      boost::input_iterator_archetype<double>(),
106      boost::output_iterator_archetype<double>(dummy_cons));
107  }
108}
109
110
111void test_column_normalize(test::Suite& suite)
112{
113  using namespace normalizer;
114  suite.err() << "Testing ColumnNormalizer\n";
115 
116  utility::Matrix m(2,2);
117  m(0,0) = 0;
118  m(0,1) = 10;
119  m(1,0) = 2;
120  m(1,1) = 4;
121  ColumnNormalizer<Centralizer<> > qn;
122  qn(m, m);
123  suite.err() << "Testing m(0,0)\n";
124  suite.add(suite.equal(m(0,0), -1));
125  suite.err() << "Testing m(0,1)\n";
126  suite.add(suite.equal(m(0,1), 3));
127  suite.err() << "Testing m(1,0)\n";
128  suite.add(suite.equal(m(1,0), 1));
129  suite.err() << "Testing m(1,1)\n";
130  suite.add(suite.equal(m(1,1), -3));
131}
132
133
134void test_qquantile_normalize(test::Suite& suite)
135{
136  using namespace normalizer;
137
138  suite.err() << "Testing qQuantileNormalizer\n";
139  std::string data(test::filename("data/normalization_test.data"));
140  if (utility::FileUtil(data.c_str()).permissions("r")) {
141    suite.add(false);
142    suite.err() << "Cannot access file " << data << '\n';
143    return;
144  }
145  std::ifstream data_stream(data.c_str());
146
147  utility::Matrix m(data_stream);
148
149  suite.err() << "testing number of parts (Q) boundary conditions\n";
150  qQuantileNormalizer(m.begin_column(0), m.end_column(0), m.rows());
151  qQuantileNormalizer(m.begin_column(0), m.end_column(0), 3);
152
153  // first column as target
154  qQuantileNormalizer qqn(m.begin_column(0), m.end_column(0) ,9); 
155  ColumnNormalizer<qQuantileNormalizer> cn(qqn);
156  utility::Matrix result(m.rows(),m.columns());
157  cn(m, result);
158
159  suite.err() << "test that result can be stored in the source matrix...";
160  cn(m,m);
161  if (suite.add(result==m))
162    suite.err() << " ok.\n";
163  else 
164    suite.err() << " failed.\n";
165
166  // Enough iteration will make all columns to have the same values as
167  // the target.
168  suite.err() << "Testing that q=matrix rows gives QuantileNormalization\n";
169  utility::Matrix m2(4,2);
170  m2(0,0) = 0; m2(0,1) = 10;
171  m2(1,0) = 2; m2(1,1) = 4;
172  m2(2,0) = 1; m2(2,1) = 0;
173  m2(3,0) = 3; m2(3,1) = 7;
174  qQuantileNormalizer qqn2(m2.begin_column(0), m2.end_column(0), m2.rows());
175  ColumnNormalizer<qQuantileNormalizer> cn2(qqn2);
176  utility::Matrix result2(m2.rows(),m2.columns());
177  cn2(m2,result2);
178  suite.add( suite.equal_fix(m2(0,0),result2(2,1),1.0e-12) &&
179             suite.equal_fix(m2(1,0),result2(3,1),1.0e-12) &&
180             suite.equal_fix(m2(2,0),result2(1,1),1.0e-12) &&
181             suite.equal_fix(m2(3,0),result2(0,1),1.0e-12) );
182}
183
184
185void test_qquantile_normalize_weighted(test::Suite& suite)
186{
187  using namespace normalizer;
188
189  suite.err() << "Testing qQuantileNormalizer weighted\n";
190
191  // test with unweighted target and source
192  std::vector<double> target;
193  target.reserve(1000);
194  while (target.size()<1000)
195    target.push_back(target.size());
196  qQuantileNormalizer qQN(target.begin(), target.end(), 4);
197  std::vector<double> source;
198  while (source.size()<10)
199    source.push_back(source.size()*10);
200  std::vector<double> result(source.size());
201 
202  qQN(source.begin(), source.end(), result.begin());
203 
204  using utility::DataWeight;
205  suite.err() << "Testing with unweighted target and weighted source\n";
206  std::vector<utility::DataWeight> src_w(source.size(), DataWeight(0, 1));
207  std::copy(source.begin(), source.end(),
208            utility::data_iterator(src_w.begin()));
209
210  std::vector<utility::DataWeight> result_w(src_w.size());
211  qQN(src_w.begin(), src_w.end(), result_w.begin());
212  suite.add(suite.equal_range(result.begin(), result.end(),
213                              utility::data_iterator(result_w.begin())));
214
215  suite.err() << "Testing with missing value in source\n";
216  // adding a missing value
217  std::vector<utility::DataWeight>::iterator MWi=src_w.begin();
218  MWi+=5;
219  src_w.insert(MWi, DataWeight(std::numeric_limits<double>::quiet_NaN(), 0.0));
220  std::vector<utility::DataWeight> result_w2(src_w.size());
221  qQN(src_w.begin(), src_w.end(), result_w2.begin());
222  // excluding missing value from comparison in suite.equal_range
223  MWi=result_w2.begin();
224  MWi+=5;
225  result_w2.erase(MWi);
226  suite.add(suite.equal_range(utility::data_iterator(result_w.begin()), 
227                              utility::data_iterator(result_w.end()),
228                              utility::data_iterator(result_w2.begin())));
229
230  suite.err() << "testing with weighted target" << std::endl;
231  std::vector<utility::DataWeight> target_w(target.size()+1, DataWeight(0, 1));
232  target_w[0] = DataWeight(5.3, 0);
233  std::copy(target.begin(), target.end(),
234            utility::data_iterator(target_w.begin()+1));
235  qQuantileNormalizer qQNw(target_w.begin(), target_w.end(), 4);
236  std::vector<utility::DataWeight> result_w3(src_w.size());
237  qQNw(src_w.begin(), src_w.end(), result_w3.begin());
238  // excluding missing value from comparison in suite.equal_range
239  MWi=result_w3.begin();
240  MWi+=5;
241  result_w3.erase(MWi);
242  suite.add(suite.equal_range(utility::data_iterator(result_w3.begin()), 
243                              utility::data_iterator(result_w3.end()),
244                              utility::data_iterator(result_w2.begin())));
245 
246}
247
248
249void test_quantile_normalize(test::Suite& suite)
250{
251  suite.err() << "Testing QuantileNormalizer\n";
252 
253  utility::Matrix m(2,2);
254  m(0,0) = 0;
255  m(0,1) = 10;
256  m(1,0) = 2;
257  m(1,1) = 4;
258  normalizer::QuantileNormalizer qn;
259  qn(m, m);
260  suite.err() << "Testing m(0,0)\n";
261  suite.add(suite.equal(m(0,0), 2));
262  suite.err() << "Testing m(0,1)\n";
263  suite.add(suite.equal(m(0,1), 6));
264  suite.err() << "Testing m(1,0)\n";
265  suite.add(suite.equal(m(1,0), 6));
266  suite.err() << "Testing m(1,1)\n";
267  suite.add(suite.equal(m(1,1), 2));
268}
269
270void test_row_normalize(test::Suite& suite)
271{
272  using namespace normalizer;
273  suite.err() << "Testing RowNormalizer\n";
274 
275  utility::Matrix m(2,3);
276  m(0,0) = 0;
277  m(0,1) = 10;
278  m(1,0) = 2;
279  m(1,1) = 4;
280  utility::Matrix m2(m);
281  m2.transpose();
282  ColumnNormalizer<Centralizer<> > cn;
283  RowNormalizer<Centralizer<> > rn;
284  cn(m, m);
285  rn(m2, m2);
286  m2.transpose();
287  suite.equal_range(m.begin(), m.end(), m2.begin());
288}
289
290void test_spearman(test::Suite& suite)
291{
292  suite.err() << "Testing Spearman\n";
293  normalizer::Spearman spearman;
294  std::vector<double> vec;
295  vec.push_back(0);
296  vec.push_back(2);
297  vec.push_back(3);
298  vec.push_back(1);
299  spearman(vec.begin(), vec.end(), vec.begin());
300  std::vector<double> correct;
301  correct.push_back(1.0/8);
302  correct.push_back(5.0/8);
303  correct.push_back(7.0/8);
304  correct.push_back(3.0/8);
305  suite.add(suite.equal_range(vec.begin(), vec.end(), correct.begin()));
306  suite.err() << "Testing Spearman with ties\n";
307  vec[1]=vec[2];
308  correct[1] = correct[2] = (correct[1]+correct[2])/2;
309  spearman(vec.begin(), vec.end(), vec.begin());
310  suite.add(suite.equal_range(vec.begin(), vec.end(), correct.begin()));
311  test_spearman_weighted(suite);
312}
313
314
315void test_gauss_normalize(test::Suite& suite)
316{
317  suite.err() << "Testing Gauss\n";
318  normalizer::Gauss gauss;
319  std::vector<double> vec;
320  vec.push_back(1);
321  gauss(vec.begin(), vec.end(), vec.begin());
322  suite.add(suite.equal(vec.front(), 0));
323  vec.push_back(1);
324  gauss(vec.begin(), vec.end(), vec.begin());
325  suite.add(suite.equal(vec.front(), -vec.back()));
326
327}
328
329void test_spearman_weighted(test::Suite& suite)
330{
331  suite.err() << "Testing Weighted Spearman\n";
332  normalizer::Spearman spearman;
333
334  suite.err() << "Testing that unity weights reproduces unweighted case\n";
335  utility::MatrixWeighted m(1,4,0,1);
336  utility::MatrixWeighted res(m.rows(), m.columns(),3.14,0);
337  m(0,0).data()=0;
338  m(0,1).data()=2;
339  m(0,2).data()=3;
340  m(0,3).data()=1;
341  std::vector<double> correct(m.columns());
342  std::vector<double> correct_w(m.columns(), 1.0);
343  std::copy(utility::data_iterator(m.begin_row(0)),
344            utility::data_iterator(m.end_row(0)),
345            correct.begin());
346  spearman(correct.begin(), correct.end(), correct.begin());
347  spearman(m.begin_row(0), m.end_row(0), res.begin_row(0));
348
349  using utility::data_iterator;
350  suite.add(suite.equal_range(data_iterator(res.begin_row(0)),
351                               data_iterator(res.end_row(0)),
352                               correct.begin()));
353  using utility::weight_iterator;
354  suite.add(suite.equal_range(weight_iterator(res.begin_row(0)),
355                               weight_iterator(res.end_row(0)),
356                               correct_w.begin()));
357
358  suite.err() << "Testing rescaling of weights\n";
359  for (size_t i=0; i<m.columns(); ++i) {
360    m(0,i).weight() *= 2;
361    correct_w[i] *= 2;
362  }   
363  spearman(m.begin_row(0), m.end_row(0), res.begin_row(0));
364  suite.add(suite.equal_range(data_iterator(res.begin_row(0)),
365                               data_iterator(res.end_row(0)),
366                               correct.begin()));
367  suite.add(suite.equal_range(weight_iterator(res.begin_row(0)),
368                               weight_iterator(res.end_row(0)),
369                               correct_w.begin()));
370
371 
372  suite.err() << "Testing case with a zero weight\n";
373  m(0,1).data() = std::numeric_limits<double>::quiet_NaN();
374  m(0,1).weight() = 0.0;
375  spearman(m.begin_row(0), m.end_row(0), res.begin_row(0));
376  suite.add(suite.equal(res(0,0).data(), 0.5/3)); 
377  suite.add(suite.equal(res(0,2).data(), 2.5/3)); 
378  suite.add(suite.equal(res(0,3).data(), 1.5/3)); 
379
380  suite.err() << "Testing case with ties\n";
381  m(0,0).data() = m(0,2).data();
382  spearman(m.begin_row(0), m.end_row(0), res.begin_row(0));
383  suite.add(suite.equal(res(0,0).data(), 2.0/3)); 
384  suite.add(suite.equal(res(0,2).data(), 2.0/3)); 
385  suite.add(suite.equal(res(0,3).data(), 0.5/3)); 
386}
387
388void test_z_score(test::Suite& suite)
389{
390  suite.err() << "Testing Zscore\n";
391  std::vector<double> vec;
392  vec.push_back(0);
393  vec.push_back(3.14);
394  normalizer::Zscore zscore;
395  zscore(vec.begin(), vec.end(), vec.begin());
396  for (size_t i=0; i<vec.size(); ++i)
397    suite.add(suite.equal(vec[i], 2.0*i-1.0));
398
399  std::vector<utility::DataWeight> vec2;
400  vec2.push_back(utility::DataWeight(1,1));
401  vec2.push_back(utility::DataWeight(2.13,0.5));
402  vec2.push_back(utility::DataWeight(2.13,0.5));
403  std::vector<utility::DataWeight> vec3(vec2.size());
404  zscore(vec2.begin(), vec2.end(), vec3.begin());
405  for (size_t i=0; i<vec2.size(); ++i)
406    suite.add(suite.equal(vec3[i].weight(), vec2[i].weight()));
407  suite.add(suite.equal(vec3[0].data(), -1.0));
408  suite.add(suite.equal(vec3[1].data(), 1.0));
409  suite.add(suite.equal(vec3[2].data(), 1.0));
410}
411
412
Note: See TracBrowser for help on using the repository browser.