source: trunk/test/normalization_test.cc @ 2147

Last change on this file since 2147 was 2147, checked in by Peter, 13 years ago

fixes #581

added archetype classes for concepts Container2D and Mutable
Container2D in namespace test. Using these to test ColumnNormalizer?
and RowNormalizer?.

  • Property svn:eol-style set to native
  • Property svn:keywords set to Id
File size: 13.3 KB
Line 
1// $Id: normalization_test.cc 2147 2010-01-17 01:43:38Z 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}
189
190
191void test_qquantile_normalize_weighted(test::Suite& suite)
192{
193  using namespace normalizer;
194
195  suite.err() << "Testing qQuantileNormalizer weighted\n";
196
197  // test with unweighted target and source
198  std::vector<double> target;
199  target.reserve(1000);
200  while (target.size()<1000)
201    target.push_back(target.size());
202  qQuantileNormalizer qQN(target.begin(), target.end(), 4);
203  std::vector<double> source;
204  while (source.size()<10)
205    source.push_back(source.size()*10);
206  std::vector<double> result(source.size());
207 
208  qQN(source.begin(), source.end(), result.begin());
209 
210  using utility::DataWeight;
211  suite.err() << "Testing with unweighted target and weighted source\n";
212  std::vector<utility::DataWeight> src_w(source.size(), DataWeight(0, 1));
213  std::copy(source.begin(), source.end(),
214            utility::data_iterator(src_w.begin()));
215
216  std::vector<utility::DataWeight> result_w(src_w.size());
217  qQN(src_w.begin(), src_w.end(), result_w.begin());
218  suite.add(suite.equal_range(result.begin(), result.end(),
219                              utility::data_iterator(result_w.begin())));
220
221  suite.err() << "Testing with missing value in source\n";
222  // adding a missing value
223  std::vector<utility::DataWeight>::iterator MWi=src_w.begin();
224  MWi+=5;
225  src_w.insert(MWi, DataWeight(std::numeric_limits<double>::quiet_NaN(), 0.0));
226  std::vector<utility::DataWeight> result_w2(src_w.size());
227  qQN(src_w.begin(), src_w.end(), result_w2.begin());
228  // excluding missing value from comparison in suite.equal_range
229  MWi=result_w2.begin();
230  MWi+=5;
231  result_w2.erase(MWi);
232  suite.add(suite.equal_range(utility::data_iterator(result_w.begin()), 
233                              utility::data_iterator(result_w.end()),
234                              utility::data_iterator(result_w2.begin())));
235
236  suite.err() << "testing with weighted target" << std::endl;
237  std::vector<utility::DataWeight> target_w(target.size()+1, DataWeight(0, 1));
238  target_w[0] = DataWeight(5.3, 0);
239  std::copy(target.begin(), target.end(),
240            utility::data_iterator(target_w.begin()+1));
241  qQuantileNormalizer qQNw(target_w.begin(), target_w.end(), 4);
242  std::vector<utility::DataWeight> result_w3(src_w.size());
243  qQNw(src_w.begin(), src_w.end(), result_w3.begin());
244  // excluding missing value from comparison in suite.equal_range
245  MWi=result_w3.begin();
246  MWi+=5;
247  result_w3.erase(MWi);
248  suite.add(suite.equal_range(utility::data_iterator(result_w3.begin()), 
249                              utility::data_iterator(result_w3.end()),
250                              utility::data_iterator(result_w2.begin())));
251 
252}
253
254
255void test_quantile_normalize(test::Suite& suite)
256{
257  suite.err() << "Testing QuantileNormalizer\n";
258 
259  utility::Matrix m(2,2);
260  m(0,0) = 0;
261  m(0,1) = 10;
262  m(1,0) = 2;
263  m(1,1) = 4;
264  normalizer::QuantileNormalizer qn;
265  qn(m, m);
266  suite.err() << "Testing m(0,0)\n";
267  suite.add(suite.equal(m(0,0), 2));
268  suite.err() << "Testing m(0,1)\n";
269  suite.add(suite.equal(m(0,1), 6));
270  suite.err() << "Testing m(1,0)\n";
271  suite.add(suite.equal(m(1,0), 6));
272  suite.err() << "Testing m(1,1)\n";
273  suite.add(suite.equal(m(1,1), 2));
274}
275
276void test_row_normalize(test::Suite& suite)
277{
278  using namespace normalizer;
279  suite.err() << "Testing RowNormalizer\n";
280 
281  utility::Matrix m(2,3);
282  m(0,0) = 0;
283  m(0,1) = 10;
284  m(1,0) = 2;
285  m(1,1) = 4;
286  utility::Matrix m2(m);
287  m2.transpose();
288  ColumnNormalizer<Centralizer<> > cn;
289  RowNormalizer<Centralizer<> > rn;
290  cn(m, m);
291  rn(m2, m2);
292  m2.transpose();
293  suite.equal_range(m.begin(), m.end(), m2.begin());
294  if (false) { // do not run compile tests
295    test::container2d_archetype<double> container2d; 
296    test::mutable_container2d_archetype<double> mutable_container2d; 
297    rn(container2d, mutable_container2d);
298  }
299}
300
301void test_spearman(test::Suite& suite)
302{
303  suite.err() << "Testing Spearman\n";
304  normalizer::Spearman spearman;
305  std::vector<double> vec;
306  vec.push_back(0);
307  vec.push_back(2);
308  vec.push_back(3);
309  vec.push_back(1);
310  spearman(vec.begin(), vec.end(), vec.begin());
311  std::vector<double> correct;
312  correct.push_back(1.0/8);
313  correct.push_back(5.0/8);
314  correct.push_back(7.0/8);
315  correct.push_back(3.0/8);
316  suite.add(suite.equal_range(vec.begin(), vec.end(), correct.begin()));
317  suite.err() << "Testing Spearman with ties\n";
318  vec[1]=vec[2];
319  correct[1] = correct[2] = (correct[1]+correct[2])/2;
320  spearman(vec.begin(), vec.end(), vec.begin());
321  suite.add(suite.equal_range(vec.begin(), vec.end(), correct.begin()));
322  test_spearman_weighted(suite);
323}
324
325
326void test_gauss_normalize(test::Suite& suite)
327{
328  suite.err() << "Testing Gauss\n";
329  normalizer::Gauss gauss;
330  std::vector<double> vec;
331  vec.push_back(1);
332  gauss(vec.begin(), vec.end(), vec.begin());
333  suite.add(suite.equal(vec.front(), 0));
334  vec.push_back(1);
335  gauss(vec.begin(), vec.end(), vec.begin());
336  suite.add(suite.equal(vec.front(), -vec.back()));
337
338}
339
340void test_spearman_weighted(test::Suite& suite)
341{
342  suite.err() << "Testing Weighted Spearman\n";
343  normalizer::Spearman spearman;
344
345  suite.err() << "Testing that unity weights reproduces unweighted case\n";
346  utility::MatrixWeighted m(1,4,0,1);
347  utility::MatrixWeighted res(m.rows(), m.columns(),3.14,0);
348  m(0,0).data()=0;
349  m(0,1).data()=2;
350  m(0,2).data()=3;
351  m(0,3).data()=1;
352  std::vector<double> correct(m.columns());
353  std::vector<double> correct_w(m.columns(), 1.0);
354  std::copy(utility::data_iterator(m.begin_row(0)),
355            utility::data_iterator(m.end_row(0)),
356            correct.begin());
357  spearman(correct.begin(), correct.end(), correct.begin());
358  spearman(m.begin_row(0), m.end_row(0), res.begin_row(0));
359
360  using utility::data_iterator;
361  suite.add(suite.equal_range(data_iterator(res.begin_row(0)),
362                               data_iterator(res.end_row(0)),
363                               correct.begin()));
364  using utility::weight_iterator;
365  suite.add(suite.equal_range(weight_iterator(res.begin_row(0)),
366                               weight_iterator(res.end_row(0)),
367                               correct_w.begin()));
368
369  suite.err() << "Testing rescaling of weights\n";
370  for (size_t i=0; i<m.columns(); ++i) {
371    m(0,i).weight() *= 2;
372    correct_w[i] *= 2;
373  }   
374  spearman(m.begin_row(0), m.end_row(0), res.begin_row(0));
375  suite.add(suite.equal_range(data_iterator(res.begin_row(0)),
376                               data_iterator(res.end_row(0)),
377                               correct.begin()));
378  suite.add(suite.equal_range(weight_iterator(res.begin_row(0)),
379                               weight_iterator(res.end_row(0)),
380                               correct_w.begin()));
381
382 
383  suite.err() << "Testing case with a zero weight\n";
384  m(0,1).data() = std::numeric_limits<double>::quiet_NaN();
385  m(0,1).weight() = 0.0;
386  spearman(m.begin_row(0), m.end_row(0), res.begin_row(0));
387  suite.add(suite.equal(res(0,0).data(), 0.5/3)); 
388  suite.add(suite.equal(res(0,2).data(), 2.5/3)); 
389  suite.add(suite.equal(res(0,3).data(), 1.5/3)); 
390
391  suite.err() << "Testing case with ties\n";
392  m(0,0).data() = m(0,2).data();
393  spearman(m.begin_row(0), m.end_row(0), res.begin_row(0));
394  suite.add(suite.equal(res(0,0).data(), 2.0/3)); 
395  suite.add(suite.equal(res(0,2).data(), 2.0/3)); 
396  suite.add(suite.equal(res(0,3).data(), 0.5/3)); 
397}
398
399void test_z_score(test::Suite& suite)
400{
401  suite.err() << "Testing Zscore\n";
402  std::vector<double> vec;
403  vec.push_back(0);
404  vec.push_back(3.14);
405  normalizer::Zscore zscore;
406  zscore(vec.begin(), vec.end(), vec.begin());
407  for (size_t i=0; i<vec.size(); ++i)
408    suite.add(suite.equal(vec[i], 2.0*i-1.0));
409
410  std::vector<utility::DataWeight> vec2;
411  vec2.push_back(utility::DataWeight(1,1));
412  vec2.push_back(utility::DataWeight(2.13,0.5));
413  vec2.push_back(utility::DataWeight(2.13,0.5));
414  std::vector<utility::DataWeight> vec3(vec2.size());
415  zscore(vec2.begin(), vec2.end(), vec3.begin());
416  for (size_t i=0; i<vec2.size(); ++i)
417    suite.add(suite.equal(vec3[i].weight(), vec2[i].weight()));
418  suite.add(suite.equal(vec3[0].data(), -1.0));
419  suite.add(suite.equal(vec3[1].data(), 1.0));
420  suite.add(suite.equal(vec3[2].data(), 1.0));
421}
422
423
Note: See TracBrowser for help on using the repository browser.