1 | #ifndef _theplu_yat_normalizer_qquantile_normalizer_ |
---|
2 | #define _theplu_yat_normalizer_qquantile_normalizer_ |
---|
3 | |
---|
4 | /* |
---|
5 | Copyright (C) 2009 Jari Häkkinen, Peter Johansson |
---|
6 | |
---|
7 | This file is part of the yat library, http://dev.thep.lu.se/yat |
---|
8 | |
---|
9 | The yat library is free software; you can redistribute it and/or |
---|
10 | modify it under the terms of the GNU General Public License as |
---|
11 | published by the Free Software Foundation; either version 3 of the |
---|
12 | License, or (at your option) any later version. |
---|
13 | |
---|
14 | The yat library is distributed in the hope that it will be useful, |
---|
15 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
---|
16 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
---|
17 | General Public License for more details. |
---|
18 | |
---|
19 | You should have received a copy of the GNU General Public License |
---|
20 | along with yat. If not, see <http://www.gnu.org/licenses/>. |
---|
21 | */ |
---|
22 | |
---|
23 | #include "yat/regression/CSplineInterpolation.h" |
---|
24 | #include "yat/utility/DataIterator.h" |
---|
25 | #include "yat/utility/DataWeight.h" |
---|
26 | #include "yat/utility/iterator_traits.h" |
---|
27 | #include "yat/utility/Vector.h" |
---|
28 | #include "yat/utility/WeightIterator.h" |
---|
29 | #include "yat/utility/yat_assert.h" |
---|
30 | |
---|
31 | #include <algorithm> |
---|
32 | #include <iterator> |
---|
33 | #include <stdexcept> |
---|
34 | #include <vector> |
---|
35 | |
---|
36 | namespace theplu { |
---|
37 | namespace yat { |
---|
38 | namespace utility { |
---|
39 | class VectorBase; |
---|
40 | } |
---|
41 | namespace normalizer { |
---|
42 | |
---|
43 | /** |
---|
44 | \brief Perform Q-quantile normalization |
---|
45 | |
---|
46 | After a Q-quantile normalization each column has approximately |
---|
47 | the same distribution of data (the Q-quantiles are the |
---|
48 | same). Also, within each column the rank of an element is not |
---|
49 | changed. |
---|
50 | |
---|
51 | There is currently no weighted version of qQuantileNormalizer |
---|
52 | |
---|
53 | The normalization goes like this |
---|
54 | - Data is not assumed to be sorted. |
---|
55 | - Partition sorted target data in N parts. N must be 3 larger |
---|
56 | because of requirements from the underlying cspline fit |
---|
57 | - Calculate the arithmetic mean for each part, the mean is |
---|
58 | assigned to the mid point of each part. |
---|
59 | - Do the above for the data to be tranformed (called source |
---|
60 | here). |
---|
61 | - For each part, calculate the difference between the target and |
---|
62 | the source. Now we have N differences d_i with associated rank |
---|
63 | (midpoint of each part). |
---|
64 | - Create a cubic spline fit to this difference vector d. The |
---|
65 | resulting curve is used to recalculate all column values. |
---|
66 | - Use the cubic spline fit for values within the cubic spline |
---|
67 | fit range [midpoint 1st part, midpoint last part]. |
---|
68 | - For data outside the cubic spline fit use linear |
---|
69 | extrapolation, i.e., a constant shift. d_first for points |
---|
70 | below fit range, and d_last for points above fit range. |
---|
71 | |
---|
72 | \since New in yat 0.5 |
---|
73 | */ |
---|
74 | class qQuantileNormalizer |
---|
75 | { |
---|
76 | public: |
---|
77 | /** |
---|
78 | \brief Documentation please. |
---|
79 | |
---|
80 | \a Q is the number of parts and must be within \f$ [3,N] \f$ |
---|
81 | where \f$ N \f$ is the total number of data points in the |
---|
82 | target. However, if \f$ N \f$ is larger than the number of points |
---|
83 | in the data to be normalized the behaviour of the code is |
---|
84 | undefined. Keep \f$ N \f$ equal to or less than the smallest |
---|
85 | number of data points in the target or each data set to be |
---|
86 | normalized against a given target. The lower bound of three is |
---|
87 | due to restrictions in the cspline fit utilized in the |
---|
88 | normalization. |
---|
89 | */ |
---|
90 | template<typename BidirectionalIterator> |
---|
91 | qQuantileNormalizer(BidirectionalIterator first, BidirectionalIterator last, |
---|
92 | unsigned int Q); |
---|
93 | |
---|
94 | /** |
---|
95 | \brief perform the Q-quantile normalization. |
---|
96 | |
---|
97 | It is possible to normalize "in place"; it is permissible for |
---|
98 | \a matrix and \a result to reference the same Matrix. |
---|
99 | |
---|
100 | \note dimensions of \a matrix and \a result must match. |
---|
101 | */ |
---|
102 | template<typename RandomAccessIterator1, typename RandomAccessIterator2> |
---|
103 | void operator()(RandomAccessIterator1 first, RandomAccessIterator1 last, |
---|
104 | RandomAccessIterator2 result) const; |
---|
105 | |
---|
106 | private: |
---|
107 | |
---|
108 | /** |
---|
109 | \brief Partition a vector of data into equal sizes. |
---|
110 | |
---|
111 | The class also calculates the average of each part and assigns |
---|
112 | the average to the mid point of each part. The midpoint is a |
---|
113 | double, i.e., it is not forced to be an integer index. |
---|
114 | */ |
---|
115 | class Partitioner |
---|
116 | { |
---|
117 | public: |
---|
118 | /** |
---|
119 | \brief Create the partition and perform required calculations. |
---|
120 | */ |
---|
121 | template<typename BidirectionalIterator> |
---|
122 | Partitioner(BidirectionalIterator first, BidirectionalIterator last, |
---|
123 | unsigned int N); |
---|
124 | |
---|
125 | /** |
---|
126 | \brief Return the averages for each part. |
---|
127 | |
---|
128 | \return The average vector. |
---|
129 | */ |
---|
130 | const utility::Vector& averages(void) const; |
---|
131 | |
---|
132 | /** |
---|
133 | \brief Return the mid point for each partition. |
---|
134 | |
---|
135 | \return The index vector. |
---|
136 | */ |
---|
137 | const utility::Vector& index(void) const; |
---|
138 | |
---|
139 | /** |
---|
140 | \return The number of parts. |
---|
141 | */ |
---|
142 | size_t size(void) const; |
---|
143 | |
---|
144 | private: |
---|
145 | // unweighted "constructor" |
---|
146 | template<typename Iterator> |
---|
147 | void build(Iterator first, Iterator last, unsigned int N, |
---|
148 | utility::unweighted_iterator_tag); |
---|
149 | // weighted "constructor" |
---|
150 | template<typename Iterator> |
---|
151 | void build(Iterator first, Iterator last, unsigned int N, |
---|
152 | utility::weighted_iterator_tag); |
---|
153 | void init(const utility::VectorBase&, unsigned int N); |
---|
154 | void init(const std::vector<utility::DataWeight>&, unsigned int N); |
---|
155 | |
---|
156 | utility::Vector average_; |
---|
157 | utility::Vector index_; |
---|
158 | }; |
---|
159 | |
---|
160 | |
---|
161 | Partitioner target_; |
---|
162 | |
---|
163 | // unweighted version |
---|
164 | template<typename RandomAccessIterator1, typename RandomAccessIterator2> |
---|
165 | void normalize(const Partitioner& source,RandomAccessIterator1 first, |
---|
166 | RandomAccessIterator1 last, RandomAccessIterator2 result, |
---|
167 | utility::unweighted_iterator_tag tag) const; |
---|
168 | |
---|
169 | // weighted version |
---|
170 | // copy weights and apply unweighted version on data part |
---|
171 | template<typename RandomAccessIterator1, typename RandomAccessIterator2> |
---|
172 | void normalize(const Partitioner& source,RandomAccessIterator1 first, |
---|
173 | RandomAccessIterator1 last, RandomAccessIterator2 result, |
---|
174 | utility::weighted_iterator_tag tag) const; |
---|
175 | }; |
---|
176 | |
---|
177 | |
---|
178 | // template implementations |
---|
179 | |
---|
180 | template<typename BidirectionalIterator> |
---|
181 | qQuantileNormalizer::qQuantileNormalizer(BidirectionalIterator first, |
---|
182 | BidirectionalIterator last, |
---|
183 | unsigned int Q) |
---|
184 | : target_(Partitioner(first, last, Q)) |
---|
185 | { |
---|
186 | utility::yat_assert<std::runtime_error>(Q>2, |
---|
187 | "qQuantileNormalizer: Q too small"); |
---|
188 | } |
---|
189 | |
---|
190 | |
---|
191 | template<typename RandomAccessIterator1, typename RandomAccessIterator2> |
---|
192 | void qQuantileNormalizer::operator()(RandomAccessIterator1 first, |
---|
193 | RandomAccessIterator1 last, |
---|
194 | RandomAccessIterator2 result) const |
---|
195 | { |
---|
196 | Partitioner source(first, last, target_.size()); |
---|
197 | typename utility::weighted_iterator_traits<RandomAccessIterator2>::type tag; |
---|
198 | normalize(source, first, last, result, tag); |
---|
199 | } |
---|
200 | |
---|
201 | |
---|
202 | template<typename RandomAccessIterator1, typename RandomAccessIterator2> |
---|
203 | void |
---|
204 | qQuantileNormalizer::normalize(const qQuantileNormalizer::Partitioner& source, |
---|
205 | RandomAccessIterator1 first, |
---|
206 | RandomAccessIterator1 last, |
---|
207 | RandomAccessIterator2 result, |
---|
208 | utility::unweighted_iterator_tag tag) const |
---|
209 | { |
---|
210 | utility::check_iterator_is_unweighted(first); |
---|
211 | utility::check_iterator_is_unweighted(result); |
---|
212 | size_t N = last-first; |
---|
213 | utility::yat_assert<std::runtime_error> |
---|
214 | (N >= target_.size(), "qQuantileNormalizer: Input range too small"); |
---|
215 | |
---|
216 | std::vector<size_t> sorted_index(last-first); |
---|
217 | utility::sort_index(first, last, sorted_index); |
---|
218 | |
---|
219 | utility::Vector diff(source.averages()); |
---|
220 | diff-=target_.averages(); |
---|
221 | const utility::Vector& idx=target_.index(); |
---|
222 | regression::CSplineInterpolation cspline(idx,diff); |
---|
223 | |
---|
224 | // linear interpolation for first part, i.e., use first diff for |
---|
225 | // all points in the first part. |
---|
226 | size_t start=0; |
---|
227 | size_t end=static_cast<unsigned int>(idx(0)); |
---|
228 | // The first condition below takes care of limiting case number |
---|
229 | // of parts approximately equal to the number of matrix rows and |
---|
230 | // the second condition makes sure that index is large enough |
---|
231 | // when using cspline below ... the static cast above takes the |
---|
232 | // floor whereas we want to take the "roof" forcing next index |
---|
233 | // range to be within interpolation range for the cspline. |
---|
234 | if ((end==0) || (end<idx(0))) |
---|
235 | ++end; |
---|
236 | for (size_t row=start; row<end; ++row) { |
---|
237 | size_t srow=sorted_index[row]; |
---|
238 | result[srow] = first[srow] - diff(0); |
---|
239 | } |
---|
240 | |
---|
241 | // cspline interpolation for all data between the mid points of |
---|
242 | // the first and last part |
---|
243 | start=end; |
---|
244 | end=static_cast<unsigned int>(idx(target_.size()-1)); |
---|
245 | // take care of limiting case number of parts approximately |
---|
246 | // equal to the number of matrix rows |
---|
247 | if (end==(static_cast<size_t>(N-1)) ) |
---|
248 | --end; |
---|
249 | for (size_t row=start; row<=end; ++row) { |
---|
250 | size_t srow=sorted_index[row]; |
---|
251 | result[srow] = first[srow] - cspline.evaluate(row) ; |
---|
252 | } |
---|
253 | |
---|
254 | // linear interpolation for last part, i.e., use last diff for |
---|
255 | // all points in the last part. |
---|
256 | start=end+1; |
---|
257 | end=N; |
---|
258 | for (size_t row=start; row<end; ++row) { |
---|
259 | size_t srow=sorted_index[row]; |
---|
260 | result[srow] = first[srow] - diff(diff.size()-1); |
---|
261 | } |
---|
262 | } |
---|
263 | |
---|
264 | |
---|
265 | template<typename RandomAccessIterator1, typename RandomAccessIterator2> |
---|
266 | void qQuantileNormalizer::normalize(const Partitioner& source, |
---|
267 | RandomAccessIterator1 first, |
---|
268 | RandomAccessIterator1 last, |
---|
269 | RandomAccessIterator2 result, |
---|
270 | utility::weighted_iterator_tag tag) const |
---|
271 | { |
---|
272 | // copy the weights |
---|
273 | std::copy(utility::weight_iterator<RandomAccessIterator1>(first), |
---|
274 | utility::weight_iterator<RandomAccessIterator1>(last), |
---|
275 | utility::weight_iterator<RandomAccessIterator2>(result)); |
---|
276 | // apply algorithm on data part of range |
---|
277 | normalize(source, utility::data_iterator<RandomAccessIterator1>(first), |
---|
278 | utility::data_iterator<RandomAccessIterator1>(last), |
---|
279 | utility::data_iterator<RandomAccessIterator2>(result), |
---|
280 | utility::unweighted_iterator_tag()); |
---|
281 | } |
---|
282 | |
---|
283 | |
---|
284 | template<typename BidirectionalIterator> |
---|
285 | qQuantileNormalizer::Partitioner::Partitioner(BidirectionalIterator first, |
---|
286 | BidirectionalIterator last, |
---|
287 | unsigned int N) |
---|
288 | : average_(utility::Vector(N)), index_(utility::Vector(N)) |
---|
289 | { |
---|
290 | typedef typename |
---|
291 | utility::weighted_iterator_traits<BidirectionalIterator>::type tag; |
---|
292 | build(first, last, N, tag()); |
---|
293 | |
---|
294 | } |
---|
295 | |
---|
296 | |
---|
297 | template<typename Iterator> |
---|
298 | void qQuantileNormalizer::Partitioner::build(Iterator first, Iterator last, |
---|
299 | unsigned int N, |
---|
300 | utility::unweighted_iterator_tag) |
---|
301 | { |
---|
302 | utility::Vector vec(std::distance(first, last)); |
---|
303 | std::copy(first, last, vec.begin()); |
---|
304 | std::sort(vec.begin(), vec.end()); |
---|
305 | init(vec, N); |
---|
306 | } |
---|
307 | |
---|
308 | |
---|
309 | template<typename Iterator> |
---|
310 | void qQuantileNormalizer::Partitioner::build(Iterator first, |
---|
311 | Iterator last, unsigned int N, |
---|
312 | utility::weighted_iterator_tag) |
---|
313 | { |
---|
314 | std::vector<utility::DataWeight> vec; |
---|
315 | vec.reserve(std::distance(first, last)); |
---|
316 | std::back_insert_iterator<std::vector<utility::DataWeight> > inserter(vec); |
---|
317 | std::copy(first, last, inserter); |
---|
318 | std::sort(vec.begin(), vec.end()); |
---|
319 | init(vec, N); |
---|
320 | } |
---|
321 | |
---|
322 | |
---|
323 | }}} // end of namespace normalizer, yat and thep |
---|
324 | |
---|
325 | #endif |
---|