source: trunk/test/normalization_test.cc @ 1773

Last change on this file since 1773 was 1773, checked in by Peter, 12 years ago

adding a test to check that qQN give same result with weighted source as with unweighted source. The test failed, which implies weighted and unweighted method are not in synch. refs #478

  • Property svn:eol-style set to native
  • Property svn:keywords set to Id
File size: 11.1 KB
Line 
1// $Id: normalization_test.cc 1773 2009-02-05 12:56:03Z 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 <fstream>
40#include <limits>
41#include <vector>
42
43using namespace theplu::yat;
44void test_centralizer(test::Suite&);
45void test_column_normalize(test::Suite&);
46void test_gauss_normalize(test::Suite&);
47void test_qquantile_normalize(test::Suite&);
48void test_qquantile_normalize_weighted(test::Suite&);
49void test_quantile_normalize(test::Suite&);
50void test_row_normalize(test::Suite&);
51void test_spearman(test::Suite&);
52void test_spearman_weighted(test::Suite&);
53void test_z_score(test::Suite&);
54
55int main(int argc, char* argv[])
56{ 
57  test::Suite suite(argc, argv);
58  suite.err() << "testing normalizations ... " << std::endl;
59
60  test_centralizer(suite);
61  test_column_normalize(suite);
62  test_qquantile_normalize(suite);
63  test_qquantile_normalize_weighted(suite);
64  test_quantile_normalize(suite);
65  test_gauss_normalize(suite);
66  test_row_normalize(suite);
67  test_spearman(suite);
68  test_z_score(suite);
69
70  return suite.return_value();
71}
72
73
74void test_centralizer(test::Suite& suite)
75{
76  suite.err() << "Testing Centralizer\n";
77  std::vector<double> vec;
78  vec.push_back(1);
79  vec.push_back(2);
80  vec.push_back(3);
81  normalizer::Centralizer<> c;
82  c(vec.begin(), vec.end(), vec.begin());
83  for (size_t i=0; i<vec.size(); ++i)
84    suite.add(suite.equal(vec[i], static_cast<double>(i)-1.0));
85
86  std::vector<utility::DataWeight> vec2;
87  vec2.push_back(utility::DataWeight(1,1));
88  vec2.push_back(utility::DataWeight(2,0.5));
89  vec2.push_back(utility::DataWeight(2,0.5));
90  std::vector<utility::DataWeight> vec3(vec2.size());
91  c(vec2.begin(), vec2.end(), vec3.begin());
92  for (size_t i=0; i<vec2.size(); ++i)
93    suite.add(suite.equal(vec3[i].weight(), vec2[i].weight()));
94  suite.add(suite.equal(vec3[0].data(), -0.5));
95  suite.add(suite.equal(vec3[1].data(), 0.5));
96  suite.add(suite.equal(vec3[2].data(), 0.5));
97
98}
99
100
101void test_column_normalize(test::Suite& suite)
102{
103  using namespace normalizer;
104  suite.err() << "Testing ColumnNormalizer\n";
105 
106  utility::Matrix m(2,2);
107  m(0,0) = 0;
108  m(0,1) = 10;
109  m(1,0) = 2;
110  m(1,1) = 4;
111  ColumnNormalizer<Centralizer<> > qn;
112  qn(m, m);
113  suite.err() << "Testing m(0,0)\n";
114  suite.add(suite.equal(m(0,0), -1));
115  suite.err() << "Testing m(0,1)\n";
116  suite.add(suite.equal(m(0,1), 3));
117  suite.err() << "Testing m(1,0)\n";
118  suite.add(suite.equal(m(1,0), 1));
119  suite.err() << "Testing m(1,1)\n";
120  suite.add(suite.equal(m(1,1), -3));
121}
122
123
124void test_qquantile_normalize(test::Suite& suite)
125{
126  using namespace normalizer;
127
128  suite.err() << "Testing qQuantileNormalizer\n";
129  std::string data(test::filename("data/normalization_test.data"));
130  if (utility::FileUtil(data.c_str()).permissions("r")) {
131    suite.add(false);
132    suite.err() << "Cannot access file " << data << '\n';
133    return;
134  }
135  std::ifstream data_stream(data.c_str());
136
137  utility::Matrix m(data_stream);
138
139  suite.err() << "testing number of parts (Q) boundary conditions\n";
140  qQuantileNormalizer(m.begin_column(0), m.end_column(0), m.rows());
141  qQuantileNormalizer(m.begin_column(0), m.end_column(0), 3);
142
143  // first column as target
144  qQuantileNormalizer qqn(m.begin_column(0), m.end_column(0) ,9); 
145  ColumnNormalizer<qQuantileNormalizer> cn(qqn);
146  utility::Matrix result(m.rows(),m.columns());
147  cn(m, result);
148
149  suite.err() << "test that result can be stored in the source matrix\n";
150  cn(m,m);
151  suite.add(result==m);
152
153  // Enough iteration will make all columns to have the same values as
154  // the target.
155  suite.err() << "Testing that q=matrix rows gives QuantileNormalization\n";
156  utility::Matrix m2(4,2);
157  m2(0,0) = 0; m2(0,1) = 10;
158  m2(1,0) = 2; m2(1,1) = 4;
159  m2(2,0) = 1; m2(2,1) = 0;
160  m2(3,0) = 3; m2(3,1) = 7;
161  qQuantileNormalizer qqn2(m2.begin_column(0), m2.end_column(0), m2.rows());
162  ColumnNormalizer<qQuantileNormalizer> cn2(qqn2);
163  utility::Matrix result2(m2.rows(),m2.columns());
164  cn2(m2,result2);
165  suite.add( suite.equal_fix(m2(0,0),result2(2,1),1.0e-12) &&
166             suite.equal_fix(m2(1,0),result2(3,1),1.0e-12) &&
167             suite.equal_fix(m2(2,0),result2(1,1),1.0e-12) &&
168             suite.equal_fix(m2(3,0),result2(0,1),1.0e-12) );
169}
170
171
172void test_qquantile_normalize_weighted(test::Suite& suite)
173{
174  using namespace normalizer;
175
176  suite.err() << "Testing qQuantileNormalizer weighted\n";
177
178  // test with unweighted target and source
179  std::vector<double> target;
180  while (target.size()<1000)
181    target.push_back(target.size());
182  qQuantileNormalizer qQN(target.begin(), target.end(), 4);
183  std::vector<double> source;
184  while (source.size()<10)
185    source.push_back(source.size()*10);
186  qQN(source.begin(), source.end(), source.begin());
187 
188  // test with unweighted target and weighted source
189  std::vector<utility::DataWeight> src_w(source.size());
190  std::copy(source.begin(), source.end(),
191            utility::data_iterator(src_w.begin()));
192  std::fill(utility::weight_iterator(src_w.begin()),
193            utility::weight_iterator(src_w.end()), 1.0);
194
195  qQN(src_w.begin(), src_w.end(), src_w.begin());
196  // FIXME: this test fails
197  suite.xadd(suite.equal_range(source.begin(), source.end(),
198                              utility::data_iterator(src_w.begin())));
199}
200
201
202void test_quantile_normalize(test::Suite& suite)
203{
204  suite.err() << "Testing QuantileNormalizer\n";
205 
206  utility::Matrix m(2,2);
207  m(0,0) = 0;
208  m(0,1) = 10;
209  m(1,0) = 2;
210  m(1,1) = 4;
211  normalizer::QuantileNormalizer qn;
212  qn(m, m);
213  suite.err() << "Testing m(0,0)\n";
214  suite.add(suite.equal(m(0,0), 2));
215  suite.err() << "Testing m(0,1)\n";
216  suite.add(suite.equal(m(0,1), 6));
217  suite.err() << "Testing m(1,0)\n";
218  suite.add(suite.equal(m(1,0), 6));
219  suite.err() << "Testing m(1,1)\n";
220  suite.add(suite.equal(m(1,1), 2));
221}
222
223void test_row_normalize(test::Suite& suite)
224{
225  using namespace normalizer;
226  suite.err() << "Testing RowNormalizer\n";
227 
228  utility::Matrix m(2,3);
229  m(0,0) = 0;
230  m(0,1) = 10;
231  m(1,0) = 2;
232  m(1,1) = 4;
233  utility::Matrix m2(m);
234  m2.transpose();
235  ColumnNormalizer<Centralizer<> > cn;
236  RowNormalizer<Centralizer<> > rn;
237  cn(m, m);
238  rn(m2, m2);
239  m2.transpose();
240  suite.equal_range(m.begin(), m.end(), m2.begin());
241}
242
243void test_spearman(test::Suite& suite)
244{
245  suite.err() << "Testing Spearman\n";
246  normalizer::Spearman spearman;
247  std::vector<double> vec;
248  vec.push_back(0);
249  vec.push_back(2);
250  vec.push_back(3);
251  vec.push_back(1);
252  spearman(vec.begin(), vec.end(), vec.begin());
253  std::vector<double> correct;
254  correct.push_back(1.0/8);
255  correct.push_back(5.0/8);
256  correct.push_back(7.0/8);
257  correct.push_back(3.0/8);
258  suite.add(suite.equal_range(vec.begin(), vec.end(), correct.begin()));
259  suite.err() << "Testing Spearman with ties\n";
260  vec[1]=vec[2];
261  correct[1] = correct[2] = (correct[1]+correct[2])/2;
262  spearman(vec.begin(), vec.end(), vec.begin());
263  suite.add(suite.equal_range(vec.begin(), vec.end(), correct.begin()));
264  test_spearman_weighted(suite);
265}
266
267
268void test_gauss_normalize(test::Suite& suite)
269{
270  suite.err() << "Testing Gauss\n";
271  normalizer::Gauss gauss;
272  std::vector<double> vec;
273  vec.push_back(1);
274  gauss(vec.begin(), vec.end(), vec.begin());
275  suite.add(suite.equal(vec.front(), 0));
276  vec.push_back(1);
277  gauss(vec.begin(), vec.end(), vec.begin());
278  suite.add(suite.equal(vec.front(), -vec.back()));
279
280}
281
282void test_spearman_weighted(test::Suite& suite)
283{
284  suite.err() << "Testing Weighted Spearman\n";
285  normalizer::Spearman spearman;
286
287  suite.err() << "Testing that unity weights reproduces unweighted case\n";
288  utility::MatrixWeighted m(1,4,0,1);
289  utility::MatrixWeighted res(m.rows(), m.columns(),3.14,0);
290  m(0,0).data()=0;
291  m(0,1).data()=2;
292  m(0,2).data()=3;
293  m(0,3).data()=1;
294  std::vector<double> correct(m.columns());
295  std::vector<double> correct_w(m.columns(), 1.0);
296  std::copy(utility::data_iterator(m.begin_row(0)),
297            utility::data_iterator(m.end_row(0)),
298            correct.begin());
299  spearman(correct.begin(), correct.end(), correct.begin());
300  spearman(m.begin_row(0), m.end_row(0), res.begin_row(0));
301
302  using utility::data_iterator;
303  suite.add(suite.equal_range(data_iterator(res.begin_row(0)),
304                               data_iterator(res.end_row(0)),
305                               correct.begin()));
306  using utility::weight_iterator;
307  suite.add(suite.equal_range(weight_iterator(res.begin_row(0)),
308                               weight_iterator(res.end_row(0)),
309                               correct_w.begin()));
310
311  suite.err() << "Testing rescaling of weights\n";
312  for (size_t i=0; i<m.columns(); ++i) {
313    m(0,i).weight() *= 2;
314    correct_w[i] *= 2;
315  }   
316  spearman(m.begin_row(0), m.end_row(0), res.begin_row(0));
317  suite.add(suite.equal_range(data_iterator(res.begin_row(0)),
318                               data_iterator(res.end_row(0)),
319                               correct.begin()));
320  suite.add(suite.equal_range(weight_iterator(res.begin_row(0)),
321                               weight_iterator(res.end_row(0)),
322                               correct_w.begin()));
323
324 
325  suite.err() << "Testing case with a zero weight\n";
326  m(0,1).data() = std::numeric_limits<double>::quiet_NaN();
327  m(0,1).weight() = 0.0;
328  spearman(m.begin_row(0), m.end_row(0), res.begin_row(0));
329  suite.add(suite.equal(res(0,0).data(), 0.5/3)); 
330  suite.add(suite.equal(res(0,2).data(), 2.5/3)); 
331  suite.add(suite.equal(res(0,3).data(), 1.5/3)); 
332
333  suite.err() << "Testing case with ties\n";
334  m(0,0).data() = m(0,2).data();
335  spearman(m.begin_row(0), m.end_row(0), res.begin_row(0));
336  suite.add(suite.equal(res(0,0).data(), 2.0/3)); 
337  suite.add(suite.equal(res(0,2).data(), 2.0/3)); 
338  suite.add(suite.equal(res(0,3).data(), 0.5/3)); 
339}
340
341void test_z_score(test::Suite& suite)
342{
343  suite.err() << "Testing Zscore\n";
344  std::vector<double> vec;
345  vec.push_back(0);
346  vec.push_back(3.14);
347  normalizer::Zscore zscore;
348  zscore(vec.begin(), vec.end(), vec.begin());
349  for (size_t i=0; i<vec.size(); ++i)
350    suite.add(suite.equal(vec[i], 2.0*i-1.0));
351
352  std::vector<utility::DataWeight> vec2;
353  vec2.push_back(utility::DataWeight(1,1));
354  vec2.push_back(utility::DataWeight(2.13,0.5));
355  vec2.push_back(utility::DataWeight(2.13,0.5));
356  std::vector<utility::DataWeight> vec3(vec2.size());
357  zscore(vec2.begin(), vec2.end(), vec3.begin());
358  for (size_t i=0; i<vec2.size(); ++i)
359    suite.add(suite.equal(vec3[i].weight(), vec2[i].weight()));
360  suite.add(suite.equal(vec3[0].data(), -1.0));
361  suite.add(suite.equal(vec3[1].data(), 1.0));
362  suite.add(suite.equal(vec3[2].data(), 1.0));
363}
364
365
Note: See TracBrowser for help on using the repository browser.