source: trunk/test/normalization_test.cc @ 2150

Last change on this file since 2150 was 2150, checked in by Peter, 13 years ago

fixes #584

  • Property svn:eol-style set to native
  • Property svn:keywords set to Id
File size: 14.1 KB
Line 
1// $Id: normalization_test.cc 2150 2010-01-17 03:54:55Z 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  if (false) { // do not run compile tests
133    test::container2d_archetype<double> container2d; 
134    test::mutable_container2d_archetype<double> mutable_container2d; 
135    qn(container2d, mutable_container2d);
136  }
137}
138
139
140void test_qquantile_normalize(test::Suite& suite)
141{
142  using namespace normalizer;
143
144  suite.err() << "Testing qQuantileNormalizer\n";
145  std::string data(test::filename("data/normalization_test.data"));
146  if (utility::FileUtil(data.c_str()).permissions("r")) {
147    suite.add(false);
148    suite.err() << "Cannot access file " << data << '\n';
149    return;
150  }
151  std::ifstream data_stream(data.c_str());
152
153  utility::Matrix m(data_stream);
154
155  suite.err() << "testing number of parts (Q) boundary conditions\n";
156  qQuantileNormalizer(m.begin_column(0), m.end_column(0), m.rows());
157  qQuantileNormalizer(m.begin_column(0), m.end_column(0), 3);
158
159  // first column as target
160  qQuantileNormalizer qqn(m.begin_column(0), m.end_column(0) ,9); 
161  ColumnNormalizer<qQuantileNormalizer> cn(qqn);
162  utility::Matrix result(m.rows(),m.columns());
163  cn(m, result);
164
165  suite.err() << "test that result can be stored in the source matrix...";
166  cn(m,m);
167  if (suite.add(result==m))
168    suite.err() << " ok.\n";
169  else 
170    suite.err() << " failed.\n";
171
172  // Enough iteration will make all columns to have the same values as
173  // the target.
174  suite.err() << "Testing that q=matrix rows gives QuantileNormalization\n";
175  utility::Matrix m2(4,2);
176  m2(0,0) = 0; m2(0,1) = 10;
177  m2(1,0) = 2; m2(1,1) = 4;
178  m2(2,0) = 1; m2(2,1) = 0;
179  m2(3,0) = 3; m2(3,1) = 7;
180  qQuantileNormalizer qqn2(m2.begin_column(0), m2.end_column(0), m2.rows());
181  ColumnNormalizer<qQuantileNormalizer> cn2(qqn2);
182  utility::Matrix result2(m2.rows(),m2.columns());
183  cn2(m2,result2);
184  suite.add( suite.equal_fix(m2(0,0),result2(2,1),1.0e-12) &&
185             suite.equal_fix(m2(1,0),result2(3,1),1.0e-12) &&
186             suite.equal_fix(m2(2,0),result2(1,1),1.0e-12) &&
187             suite.equal_fix(m2(3,0),result2(0,1),1.0e-12) );
188  // compile test should not be run
189  if (false) {
190    qQuantileNormalizer qqn3(boost::forward_iterator_archetype<double>(),
191                             boost::forward_iterator_archetype<double>(),
192                             100);
193    qqn3(boost::random_access_iterator_archetype<double>(), 
194         boost::random_access_iterator_archetype<double>(),
195         boost::mutable_random_access_iterator_archetype<double>());
196  }
197}
198
199
200void test_qquantile_normalize_weighted(test::Suite& suite)
201{
202  using namespace normalizer;
203
204  suite.err() << "Testing qQuantileNormalizer weighted\n";
205
206  // test with unweighted target and source
207  std::vector<double> target;
208  target.reserve(1000);
209  while (target.size()<1000)
210    target.push_back(target.size());
211  qQuantileNormalizer qQN(target.begin(), target.end(), 4);
212  std::vector<double> source;
213  while (source.size()<10)
214    source.push_back(source.size()*10);
215  std::vector<double> result(source.size());
216 
217  qQN(source.begin(), source.end(), result.begin());
218 
219  using utility::DataWeight;
220  suite.err() << "Testing with unweighted target and weighted source\n";
221  std::vector<utility::DataWeight> src_w(source.size(), DataWeight(0, 1));
222  std::copy(source.begin(), source.end(),
223            utility::data_iterator(src_w.begin()));
224
225  std::vector<utility::DataWeight> result_w(src_w.size());
226  qQN(src_w.begin(), src_w.end(), result_w.begin());
227  suite.add(suite.equal_range(result.begin(), result.end(),
228                              utility::data_iterator(result_w.begin())));
229
230  suite.err() << "Testing with missing value in source\n";
231  // adding a missing value
232  std::vector<utility::DataWeight>::iterator MWi=src_w.begin();
233  MWi+=5;
234  src_w.insert(MWi, DataWeight(std::numeric_limits<double>::quiet_NaN(), 0.0));
235  std::vector<utility::DataWeight> result_w2(src_w.size());
236  qQN(src_w.begin(), src_w.end(), result_w2.begin());
237  // excluding missing value from comparison in suite.equal_range
238  MWi=result_w2.begin();
239  MWi+=5;
240  result_w2.erase(MWi);
241  suite.add(suite.equal_range(utility::data_iterator(result_w.begin()), 
242                              utility::data_iterator(result_w.end()),
243                              utility::data_iterator(result_w2.begin())));
244
245  suite.err() << "testing with weighted target" << std::endl;
246  std::vector<utility::DataWeight> target_w(target.size()+1, DataWeight(0, 1));
247  target_w[0] = DataWeight(5.3, 0);
248  std::copy(target.begin(), target.end(),
249            utility::data_iterator(target_w.begin()+1));
250  qQuantileNormalizer qQNw(target_w.begin(), target_w.end(), 4);
251  std::vector<utility::DataWeight> result_w3(src_w.size());
252  qQNw(src_w.begin(), src_w.end(), result_w3.begin());
253  // excluding missing value from comparison in suite.equal_range
254  MWi=result_w3.begin();
255  MWi+=5;
256  result_w3.erase(MWi);
257  suite.add(suite.equal_range(utility::data_iterator(result_w3.begin()), 
258                              utility::data_iterator(result_w3.end()),
259                              utility::data_iterator(result_w2.begin())));
260 
261}
262
263
264void test_quantile_normalize(test::Suite& suite)
265{
266  suite.err() << "Testing QuantileNormalizer\n";
267 
268  utility::Matrix m(2,2);
269  m(0,0) = 0;
270  m(0,1) = 10;
271  m(1,0) = 2;
272  m(1,1) = 4;
273  normalizer::QuantileNormalizer qn;
274  qn(m, m);
275  suite.err() << "Testing m(0,0)\n";
276  suite.add(suite.equal(m(0,0), 2));
277  suite.err() << "Testing m(0,1)\n";
278  suite.add(suite.equal(m(0,1), 6));
279  suite.err() << "Testing m(1,0)\n";
280  suite.add(suite.equal(m(1,0), 6));
281  suite.err() << "Testing m(1,1)\n";
282  suite.add(suite.equal(m(1,1), 2));
283}
284
285void test_row_normalize(test::Suite& suite)
286{
287  using namespace normalizer;
288  suite.err() << "Testing RowNormalizer\n";
289 
290  utility::Matrix m(2,3);
291  m(0,0) = 0;
292  m(0,1) = 10;
293  m(1,0) = 2;
294  m(1,1) = 4;
295  utility::Matrix m2(m);
296  m2.transpose();
297  ColumnNormalizer<Centralizer<> > cn;
298  RowNormalizer<Centralizer<> > rn;
299  cn(m, m);
300  rn(m2, m2);
301  m2.transpose();
302  suite.equal_range(m.begin(), m.end(), m2.begin());
303  if (false) { // do not run compile tests
304    test::container2d_archetype<double> container2d; 
305    test::mutable_container2d_archetype<double> mutable_container2d; 
306    rn(container2d, mutable_container2d);
307  }
308}
309
310void test_spearman(test::Suite& suite)
311{
312  suite.err() << "Testing Spearman\n";
313  normalizer::Spearman spearman;
314  std::vector<double> vec;
315  vec.push_back(0);
316  vec.push_back(2);
317  vec.push_back(3);
318  vec.push_back(1);
319  spearman(vec.begin(), vec.end(), vec.begin());
320  std::vector<double> correct;
321  correct.push_back(1.0/8);
322  correct.push_back(5.0/8);
323  correct.push_back(7.0/8);
324  correct.push_back(3.0/8);
325  suite.add(suite.equal_range(vec.begin(), vec.end(), correct.begin()));
326  suite.err() << "Testing Spearman with ties\n";
327  vec[1]=vec[2];
328  correct[1] = correct[2] = (correct[1]+correct[2])/2;
329  spearman(vec.begin(), vec.end(), vec.begin());
330  suite.add(suite.equal_range(vec.begin(), vec.end(), correct.begin()));
331  test_spearman_weighted(suite);
332}
333
334
335void test_gauss_normalize(test::Suite& suite)
336{
337  suite.err() << "Testing Gauss\n";
338  normalizer::Gauss gauss;
339  std::vector<double> vec;
340  vec.push_back(1);
341  gauss(vec.begin(), vec.end(), vec.begin());
342  suite.add(suite.equal(vec.front(), 0));
343  vec.push_back(1);
344  gauss(vec.begin(), vec.end(), vec.begin());
345  suite.add(suite.equal(vec.front(), -vec.back()));
346  // compile test should not be run
347  if (false) {
348    gauss(boost::random_access_iterator_archetype<double>(), 
349          boost::random_access_iterator_archetype<double>(),
350          boost::mutable_random_access_iterator_archetype<double>());
351  }
352}
353
354void test_spearman_weighted(test::Suite& suite)
355{
356  suite.err() << "Testing Weighted Spearman\n";
357  normalizer::Spearman spearman;
358
359  suite.err() << "Testing that unity weights reproduces unweighted case\n";
360  utility::MatrixWeighted m(1,4,0,1);
361  utility::MatrixWeighted res(m.rows(), m.columns(),3.14,0);
362  m(0,0).data()=0;
363  m(0,1).data()=2;
364  m(0,2).data()=3;
365  m(0,3).data()=1;
366  std::vector<double> correct(m.columns());
367  std::vector<double> correct_w(m.columns(), 1.0);
368  std::copy(utility::data_iterator(m.begin_row(0)),
369            utility::data_iterator(m.end_row(0)),
370            correct.begin());
371  spearman(correct.begin(), correct.end(), correct.begin());
372  spearman(m.begin_row(0), m.end_row(0), res.begin_row(0));
373
374  using utility::data_iterator;
375  suite.add(suite.equal_range(data_iterator(res.begin_row(0)),
376                               data_iterator(res.end_row(0)),
377                               correct.begin()));
378  using utility::weight_iterator;
379  suite.add(suite.equal_range(weight_iterator(res.begin_row(0)),
380                               weight_iterator(res.end_row(0)),
381                               correct_w.begin()));
382
383  suite.err() << "Testing rescaling of weights\n";
384  for (size_t i=0; i<m.columns(); ++i) {
385    m(0,i).weight() *= 2;
386    correct_w[i] *= 2;
387  }   
388  spearman(m.begin_row(0), m.end_row(0), res.begin_row(0));
389  suite.add(suite.equal_range(data_iterator(res.begin_row(0)),
390                               data_iterator(res.end_row(0)),
391                               correct.begin()));
392  suite.add(suite.equal_range(weight_iterator(res.begin_row(0)),
393                               weight_iterator(res.end_row(0)),
394                               correct_w.begin()));
395
396 
397  suite.err() << "Testing case with a zero weight\n";
398  m(0,1).data() = std::numeric_limits<double>::quiet_NaN();
399  m(0,1).weight() = 0.0;
400  spearman(m.begin_row(0), m.end_row(0), res.begin_row(0));
401  suite.add(suite.equal(res(0,0).data(), 0.5/3)); 
402  suite.add(suite.equal(res(0,2).data(), 2.5/3)); 
403  suite.add(suite.equal(res(0,3).data(), 1.5/3)); 
404
405  suite.err() << "Testing case with ties\n";
406  m(0,0).data() = m(0,2).data();
407  spearman(m.begin_row(0), m.end_row(0), res.begin_row(0));
408  suite.add(suite.equal(res(0,0).data(), 2.0/3)); 
409  suite.add(suite.equal(res(0,2).data(), 2.0/3)); 
410  suite.add(suite.equal(res(0,3).data(), 0.5/3)); 
411  // compile test should not be run
412  if (false) {
413    spearman(boost::random_access_iterator_archetype<double>(), 
414             boost::random_access_iterator_archetype<double>(),
415             boost::mutable_random_access_iterator_archetype<double>());
416  }
417
418}
419
420void test_z_score(test::Suite& suite)
421{
422  suite.err() << "Testing Zscore\n";
423  std::vector<double> vec;
424  vec.push_back(0);
425  vec.push_back(3.14);
426  normalizer::Zscore zscore;
427  zscore(vec.begin(), vec.end(), vec.begin());
428  for (size_t i=0; i<vec.size(); ++i)
429    suite.add(suite.equal(vec[i], 2.0*i-1.0));
430
431  std::vector<utility::DataWeight> vec2;
432  vec2.push_back(utility::DataWeight(1,1));
433  vec2.push_back(utility::DataWeight(2.13,0.5));
434  vec2.push_back(utility::DataWeight(2.13,0.5));
435  std::vector<utility::DataWeight> vec3(vec2.size());
436  zscore(vec2.begin(), vec2.end(), vec3.begin());
437  for (size_t i=0; i<vec2.size(); ++i)
438    suite.add(suite.equal(vec3[i].weight(), vec2[i].weight()));
439  suite.add(suite.equal(vec3[0].data(), -1.0));
440  suite.add(suite.equal(vec3[1].data(), 1.0));
441  suite.add(suite.equal(vec3[2].data(), 1.0));
442}
443
444
Note: See TracBrowser for help on using the repository browser.