| 1 |
/* BaitFisher (version 1.2.8) a program for designing DNA target enrichment baits |
| 2 |
* BaitFilter (version 1.0.6) a program for selecting optimal bait regions |
| 3 |
* Copyright 2013-2017 by Christoph Mayer |
| 4 |
* |
| 5 |
* This source file is part of the BaitFisher-package. |
| 6 |
* |
| 7 |
* This program is free software: you can redistribute it and/or modify |
| 8 |
* it under the terms of the GNU General Public License as published by |
| 9 |
* the Free Software Foundation, either version 3 of the License, or |
| 10 |
* (at your option) any later version. |
| 11 |
* |
| 12 |
* This program is distributed in the hope that it will be useful, |
| 13 |
* but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 14 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 15 |
* GNU General Public License for more details. |
| 16 |
* |
| 17 |
* You should have received a copy of the GNU General Public License |
| 18 |
* along with BaitFisher. If not, see <http://www.gnu.org/licenses/>. |
| 19 |
* |
| 20 |
* |
| 21 |
* For any enquiries send an Email to Christoph Mayer |
| 22 |
* [email protected] |
| 23 |
* |
| 24 |
* When publishing work that is based on the results please cite: |
| 25 |
* Mayer et al. 2016: BaitFisher: A software package for multi-species target DNA enrichment probe design |
| 26 |
* |
| 27 |
*/ |
| 28 |
#ifndef STATISTIC_FUNCTIONS_H |
| 29 |
#define STATISTIC_FUNCTIONS_H |
| 30 |
|
| 31 |
#include <iostream> |
| 32 |
#include <vector> |
| 33 |
#include <map> |
| 34 |
#include <cmath> |
| 35 |
#include <cctype> |
| 36 |
#include <iterator> |
| 37 |
#include <cstdlib> |
| 38 |
#include <algorithm> |
| 39 |
#include <climits> |
| 40 |
|
| 41 |
#define EPSS 0.00000000001; |
| 42 |
|
| 43 |
|
| 44 |
#define macro_min(x,y) ((x)<(y) ? (x) : (y)) |
| 45 |
#define macro_max(x,y) ((x)>(y) ? (x) : (y)) |
| 46 |
#define macro_mean_double(x,y) (((x)+(y))/2.0) |
| 47 |
|
| 48 |
|
| 49 |
template<typename T> |
| 50 |
size_t vec_mean_sd(const std::vector<T> &vec, |
| 51 |
double &mean, double &sd) |
| 52 |
{ |
| 53 |
size_t i, n=vec.size(); |
| 54 |
double sum_x = 0; |
| 55 |
double sum_xx = 0; |
| 56 |
|
| 57 |
if (n==0) |
| 58 |
return 0; |
| 59 |
|
| 60 |
for (i=0; i<n; ++i) |
| 61 |
{ |
| 62 |
sum_x += (double)vec[i]; |
| 63 |
sum_xx += (double)vec[i]*vec[i]; |
| 64 |
} |
| 65 |
if (n!=0) |
| 66 |
mean = sum_x/(double)n; |
| 67 |
else |
| 68 |
mean = 0; |
| 69 |
if (n>1) |
| 70 |
sd = sqrt( (sum_xx-sum_x*mean)/(double)(n-1.0) ); |
| 71 |
else |
| 72 |
sd = 0; |
| 73 |
if (sd < 0.000001) |
| 74 |
sd = 0; |
| 75 |
return n; |
| 76 |
} |
| 77 |
|
| 78 |
template<typename T> |
| 79 |
size_t vec_mean_sd(const std::vector<T> &vec, |
| 80 |
double &mean, double &sd, |
| 81 |
T &sum, T &sum_of_squares) |
| 82 |
{ |
| 83 |
size_t i, n=vec.size(); |
| 84 |
T sum_x = 0; |
| 85 |
T sum_xx = 0; |
| 86 |
T elem; |
| 87 |
|
| 88 |
if (n==0) |
| 89 |
return 0; |
| 90 |
|
| 91 |
for (i=0; i<n; ++i) |
| 92 |
{ |
| 93 |
elem = vec[i]; |
| 94 |
sum_x += elem; |
| 95 |
sum_xx += elem*elem; |
| 96 |
} |
| 97 |
mean = (double)sum_x/(double)n; |
| 98 |
if (n>1) |
| 99 |
sd = sqrt( (sum_xx-sum_x*mean)/(double)(n-1.0) ); |
| 100 |
else |
| 101 |
sd = 0; |
| 102 |
if (sd < 0.000001) |
| 103 |
sd = 0; |
| 104 |
|
| 105 |
sum = sum_x; |
| 106 |
sum_of_squares = sum_xx; |
| 107 |
|
| 108 |
return n; |
| 109 |
} |
| 110 |
|
| 111 |
// Coordinates conventions: |
| 112 |
// The first coordinate is the index of the first element in the range as 0 based index. |
| 113 |
// The second coordinate is the index of the last element in the range as 0 based index. |
| 114 |
template<typename T> |
| 115 |
void median_range(std::vector<T> &vec, size_t f, size_t l, double &median, size_t &index, bool &is_datum) |
| 116 |
{ |
| 117 |
size_t s = f+l; // This is not the length or size of the range. It is the sum of the indices. |
| 118 |
size_t m = s/2; // This is the index of the mid of the range |
| 119 |
|
| 120 |
// In order to see that the following code is correct one should look at the following examples: |
| 121 |
// vec: 0,1,2, Range: 0 to 2, which is the full vector. => s=2, m=1 which is the correct mid of the range. s is even so index=m=1, median is vec[m]=vec[1]. |
| 122 |
// vec: 0,1,2,3 Range: 1 to 3, => s=4, m=2, median = vec[2], index = 2 |
| 123 |
// vec: 0,1,2,3 Range: 3 to 3, => s=6, m=3, median = vec[3], index = 3 |
| 124 |
// vec: 0,1,2,3 Range: 0 to 3, => s=3, m=1, s is odd so median = (vec[1]+vec[2])/2, index = 1 |
| 125 |
// vec: 0,1,2,4,4 Range 1 to 4, => s=5, m=2, s is odd so median = (vec[2]+vec[3])/2, index = 2 |
| 126 |
// vec: 0,1,2,4,4 Range 2 to 4, => s=6, m=3, median = vec[3], index = 3 |
| 127 |
// vec: 0,1,2,4,4 Range 1 to 3, => s=4, m=2, median = vec[2], index = 2 |
| 128 |
|
| 129 |
// The following code simply relies on the fact that: |
| 130 |
// (i) m is obviously the middle of the range, or left of the middle if the number of elements is even. |
| 131 |
// (ii) s is even if the range has an odd number of elements and vice versa. |
| 132 |
|
| 133 |
if (s%2 == 0) |
| 134 |
{ |
| 135 |
median = vec[m]; |
| 136 |
index = m; |
| 137 |
is_datum = true; |
| 138 |
} |
| 139 |
else |
| 140 |
{ |
| 141 |
median = (vec[m]+vec[m+1])/2; |
| 142 |
index = m; |
| 143 |
is_datum = false; |
| 144 |
} |
| 145 |
} |
| 146 |
|
| 147 |
// Method 2 routines: |
| 148 |
|
| 149 |
// Prerequisite: Type T must be a type for which it makes sense to multiply it with a double to obtain a double. |
| 150 |
// Compute quartiles: |
| 151 |
// second quartile=median |
| 152 |
// Q1 and Q3 are medians of the first part and second part, where the first and second part are obtained by dividing the data at the median. |
| 153 |
template<typename T> |
| 154 |
size_t vec_median_quartile_sort_method2(std::vector<T> &vec, double &Q1, double &Q2, double &Q3) |
| 155 |
{ |
| 156 |
size_t n=vec.size(); |
| 157 |
size_t index, dummy; |
| 158 |
bool is_datum; |
| 159 |
|
| 160 |
std::sort(vec.begin(), vec.end()); |
| 161 |
|
| 162 |
median_range(vec, 0, n-1, Q2, index, is_datum); |
| 163 |
|
| 164 |
if (is_datum) |
| 165 |
{ |
| 166 |
median_range(vec, 0, index, Q1, dummy, is_datum); |
| 167 |
median_range(vec, index, n-1, Q3, dummy, is_datum); |
| 168 |
} |
| 169 |
else |
| 170 |
{ |
| 171 |
median_range(vec, 0, index, Q1, dummy, is_datum); |
| 172 |
median_range(vec, index+1, n-1, Q3, dummy, is_datum); |
| 173 |
} |
| 174 |
return n; |
| 175 |
} |
| 176 |
|
| 177 |
template<typename T> |
| 178 |
size_t vec_median_quartile_method2(std::vector<T> vec, double &Q1, double &Q2, double &Q3) |
| 179 |
{ |
| 180 |
return vec_median_quartile_sort_method2(vec, Q1, Q2, Q3); |
| 181 |
} |
| 182 |
|
| 183 |
// |
| 184 |
// Computes outlier bounds for a vector. |
| 185 |
// Side effect: vec gets sorted. More efficient than copying the vector. |
| 186 |
template<typename T> |
| 187 |
size_t vec_median_quartile_outlier_bounds_sort_method2(std::vector<T> &vec, double &Q1, double &Q2, double &Q3, double &O_lower, double &O_upper) |
| 188 |
{ |
| 189 |
int res = vec_median_quartile_sort_method2(vec, Q1, Q2, Q3); |
| 190 |
double IQR = Q3 - Q1; |
| 191 |
|
| 192 |
O_lower = Q1 - 1.5 * IQR; |
| 193 |
O_upper = Q3 + 1.5 * IQR; |
| 194 |
|
| 195 |
return res; |
| 196 |
} |
| 197 |
|
| 198 |
// Computes outlier bounds for a vector. |
| 199 |
// Same vec_median_quartile_outlier_bounds_sort but without sorting the vector. Slighly slower, since the vector has to be copied. |
| 200 |
template<typename T> |
| 201 |
size_t vec_median_quartile_outlier_bounds_method2(std::vector<T> vec, double &Q1, double &Q2, double &Q3, double &O_lower, double &O_upper) |
| 202 |
{ |
| 203 |
int res = vec_median_quartile_sort_method2(vec, Q1, Q2, Q3); |
| 204 |
double IQR = Q3 - Q1; |
| 205 |
|
| 206 |
O_lower = Q1 - 1.5 * IQR; |
| 207 |
O_upper = Q3 + 1.5 * IQR; |
| 208 |
|
| 209 |
return res; |
| 210 |
} |
| 211 |
|
| 212 |
template<typename T> |
| 213 |
void vec_mark_outlier_mehod2(std::vector<T> &vec, std::vector<bool> &outlier) |
| 214 |
{ |
| 215 |
outlier.clear(); |
| 216 |
|
| 217 |
double Q1, Q2, Q3, O_lower, O_upper; |
| 218 |
vec_median_quartile_outlier_bounds_method2(vec, Q1, Q2, Q3, O_lower, O_upper); |
| 219 |
|
| 220 |
unsigned i, N = vec.size(); |
| 221 |
outlier.reserve(N); |
| 222 |
for (i=0; i<N; ++i) |
| 223 |
{ |
| 224 |
if (vec[i] < O_lower || vec[i] > O_upper) |
| 225 |
outlier.push_back(true); |
| 226 |
else |
| 227 |
outlier.push_back(false); |
| 228 |
} |
| 229 |
std::cout << "Qi: " << Q1 << " " << Q2 << " " << Q3 << std::endl; |
| 230 |
std::cout << "Bounds: " << O_lower << " " << O_upper << std::endl; |
| 231 |
} |
| 232 |
|
| 233 |
// Method 3 routines: |
| 234 |
|
| 235 |
// Prerequisite: Type T must be a type for which it makes sense to multiply it with a double to obtain a double. |
| 236 |
// Compute quartiles: Averages of values at certain indices. |
| 237 |
// |
| 238 |
template<typename T> |
| 239 |
size_t vec_median_quartile_sort_method3(std::vector<T> &vec, double &Q1, double &Q2, double &Q3) |
| 240 |
{ |
| 241 |
size_t n=vec.size(); |
| 242 |
size_t m = n/4; |
| 243 |
size_t r = n%4; |
| 244 |
|
| 245 |
std::sort(vec.begin(), vec.end()); |
| 246 |
|
| 247 |
if (n==0) |
| 248 |
return 0; |
| 249 |
|
| 250 |
if (n==1) |
| 251 |
{ |
| 252 |
Q1 = vec[0]; |
| 253 |
Q2 = vec[0]; |
| 254 |
Q3 = vec[0]; |
| 255 |
return 1; |
| 256 |
} |
| 257 |
|
| 258 |
if (r == 0) |
| 259 |
{ |
| 260 |
Q1 = 0.5*vec[ m-1] + 0.5*vec[ m]; |
| 261 |
Q2 = 0.5*vec[2*m-1] + 0.5*vec[2*m]; |
| 262 |
Q3 = 0.5*vec[3*m-1] + 0.5*vec[3*m]; |
| 263 |
} |
| 264 |
else if (r == 1) |
| 265 |
{ |
| 266 |
// std::cerr << "m: " << m << std::endl; |
| 267 |
|
| 268 |
Q1 = 0.25*vec[ m-1] + 0.75*vec[ m]; |
| 269 |
Q2 = vec[2*m]; |
| 270 |
Q3 = 0.75*vec[ 3*m] + 0.25*vec[3*m+1]; |
| 271 |
} |
| 272 |
else if (r == 2) |
| 273 |
{ |
| 274 |
// std::cerr << "m: " << m << std::endl; |
| 275 |
|
| 276 |
Q1 = vec[m]; |
| 277 |
Q2 = 0.50*vec[2*m] + 0.50*vec[2*m+1]; |
| 278 |
Q3 = vec[3*m+1]; |
| 279 |
} |
| 280 |
else // if (r == 3) |
| 281 |
{ |
| 282 |
// std::cerr << "m: " << m << std::endl; |
| 283 |
|
| 284 |
Q1 = 0.75*vec[ m] + 0.25*vec[ m+1]; |
| 285 |
Q2 = vec[2*m+1]; |
| 286 |
Q3 = 0.25*vec[3*m+1] + 0.75*vec[3*m+2]; |
| 287 |
} |
| 288 |
return n; |
| 289 |
} |
| 290 |
|
| 291 |
template<typename T> |
| 292 |
size_t vec_median_quartile_method3(std::vector<T> vec, double &Q1, double &Q2, double &Q3) |
| 293 |
{ |
| 294 |
return vec_median_quartile_sort_method3(vec, Q1, Q2, Q3); |
| 295 |
} |
| 296 |
|
| 297 |
// Computes outlier bounds for a vector. |
| 298 |
// Side effect: vec gets sorted. More efficient than copying the vector. |
| 299 |
template<typename T> |
| 300 |
size_t vec_median_quartile_outlier_bounds_sort_method3(std::vector<T> &vec, double &Q1, double &Q2, double &Q3, double &O_lower, double &O_upper) |
| 301 |
{ |
| 302 |
int res = vec_median_quartile_sort_method3(vec, Q1, Q2, Q3); |
| 303 |
double IQR = Q3 - Q1; |
| 304 |
|
| 305 |
O_lower = Q1 - 1.5 * IQR; |
| 306 |
O_upper = Q3 + 1.5 * IQR; |
| 307 |
|
| 308 |
return res; |
| 309 |
} |
| 310 |
|
| 311 |
// Computes outlier bounds for a vector. |
| 312 |
// Same vec_median_quartile_outlier_bounds_sort but without sorting the vector. Slighly slower, since the vector has to be copied. |
| 313 |
template<typename T> |
| 314 |
size_t vec_median_quartile_outlier_bounds_method3(std::vector<T> vec, double &Q1, double &Q2, double &Q3, double &O_lower, double &O_upper) |
| 315 |
{ |
| 316 |
int res = vec_median_quartile_sort_method3(vec, Q1, Q2, Q3); |
| 317 |
double IQR = Q3 - Q1; |
| 318 |
|
| 319 |
O_lower = Q1 - 1.5 * IQR; |
| 320 |
O_upper = Q3 + 1.5 * IQR; |
| 321 |
|
| 322 |
return res; |
| 323 |
} |
| 324 |
|
| 325 |
template<typename T> |
| 326 |
void vec_mark_outlier_mehod3(std::vector<T> &vec, std::vector<bool> &outlier) |
| 327 |
{ |
| 328 |
outlier.clear(); |
| 329 |
|
| 330 |
double Q1, Q2, Q3, O_lower, O_upper; |
| 331 |
vec_median_quartile_outlier_bounds_method3(vec, Q1, Q2, Q3, O_lower, O_upper); |
| 332 |
|
| 333 |
unsigned i, N = vec.size(); |
| 334 |
outlier.reserve(N); |
| 335 |
for (i=0; i<N; ++i) |
| 336 |
{ |
| 337 |
if (vec[i] < O_lower || vec[i] > O_upper) |
| 338 |
outlier.push_back(true); |
| 339 |
else |
| 340 |
outlier.push_back(false); |
| 341 |
} |
| 342 |
std::cout << "Qi: " << Q1 << " " << Q2 << " " << Q3 << std::endl; |
| 343 |
std::cout << "Bounds: " << O_lower << " " << O_upper << std::endl; |
| 344 |
} |
| 345 |
|
| 346 |
template<typename T> |
| 347 |
size_t vec_min_max(const std::vector<T> &vec, T &min, T &max) |
| 348 |
{ |
| 349 |
size_t i, n=vec.size(); |
| 350 |
|
| 351 |
if (n==0) |
| 352 |
return 0; |
| 353 |
|
| 354 |
max = min = vec[0]; |
| 355 |
|
| 356 |
for (i=0; i<n; ++i) |
| 357 |
{ |
| 358 |
if (vec[i] < min) |
| 359 |
min = vec[i]; |
| 360 |
|
| 361 |
if (vec[i] > max) |
| 362 |
max = vec[i]; |
| 363 |
} |
| 364 |
return n; |
| 365 |
} |
| 366 |
|
| 367 |
|
| 368 |
// Determines the mean and sd for a range. |
| 369 |
template <typename T> |
| 370 |
size_t range_mean_sd(T it_beg, |
| 371 |
T it_end, |
| 372 |
double &mean, double &sd) |
| 373 |
{ |
| 374 |
double sum_x = 0; |
| 375 |
double sum_xx = 0; |
| 376 |
size_t n=0; |
| 377 |
|
| 378 |
if (it_beg == it_end) |
| 379 |
{ |
| 380 |
return 0; |
| 381 |
} |
| 382 |
|
| 383 |
while (it_end != it_beg) |
| 384 |
{ |
| 385 |
sum_x += (double)*it_beg; |
| 386 |
sum_xx += (double)*it_beg* *it_beg; |
| 387 |
++it_beg; |
| 388 |
++n; |
| 389 |
} |
| 390 |
mean = sum_x/(double)n; |
| 391 |
if (n>1) |
| 392 |
sd = sqrt( (sum_xx-sum_x*mean)/(double)(n-1.0) ); |
| 393 |
else // if (n==1) |
| 394 |
sd = 0; |
| 395 |
if (sd < 0.000001) |
| 396 |
sd = 0; |
| 397 |
return n; |
| 398 |
} |
| 399 |
|
| 400 |
|
| 401 |
template<typename T> |
| 402 |
size_t range_min_max(T it_beg, |
| 403 |
T it_end, |
| 404 |
double &min, double &max) |
| 405 |
{ |
| 406 |
size_t n=0; |
| 407 |
|
| 408 |
if (it_beg == it_end) |
| 409 |
return 0; |
| 410 |
|
| 411 |
max = min = *it_beg; |
| 412 |
|
| 413 |
while (it_end != it_beg) |
| 414 |
{ |
| 415 |
if (*it_beg< min) |
| 416 |
min = *it_beg; |
| 417 |
|
| 418 |
if (*it_beg > max) |
| 419 |
max = *it_beg; |
| 420 |
++n; |
| 421 |
++it_beg; |
| 422 |
} |
| 423 |
return n; |
| 424 |
} |
| 425 |
|
| 426 |
template <typename T> |
| 427 |
void vec_sum_sum_squares(std::vector<T> v, T& sum, T&sum_squares) |
| 428 |
{ |
| 429 |
sum=0; |
| 430 |
sum_squares=0; |
| 431 |
|
| 432 |
unsigned i, n=v.size(); |
| 433 |
|
| 434 |
for (i=0; i<n; ++i) |
| 435 |
{ |
| 436 |
sum += v[i]; |
| 437 |
sum_squares += v[i]*v[i]; |
| 438 |
} |
| 439 |
} |
| 440 |
|
| 441 |
// More functions: Single result as return value, using pointers to specify ranges. |
| 442 |
template <typename T> |
| 443 |
T minimum(T*b, T*e) |
| 444 |
{ |
| 445 |
T m; |
| 446 |
if (b>=e) |
| 447 |
return INT_MIN; |
| 448 |
m = *b; |
| 449 |
++b; |
| 450 |
while (b!=e) |
| 451 |
{ |
| 452 |
if (*b < m) |
| 453 |
{ |
| 454 |
m = *b; |
| 455 |
} |
| 456 |
++b; |
| 457 |
} |
| 458 |
return m; |
| 459 |
} |
| 460 |
|
| 461 |
template <typename T> |
| 462 |
T maximum(T*b, T*e) |
| 463 |
{ |
| 464 |
T m; |
| 465 |
if (b>=e) |
| 466 |
return INT_MIN; |
| 467 |
m = *b; |
| 468 |
++b; |
| 469 |
while (b!=e) |
| 470 |
{ |
| 471 |
if (*b > m) |
| 472 |
{ |
| 473 |
m = *b; |
| 474 |
} |
| 475 |
++b; |
| 476 |
} |
| 477 |
return m; |
| 478 |
} |
| 479 |
|
| 480 |
|
| 481 |
|
| 482 |
|
| 483 |
|
| 484 |
#endif |