[29] | 1 | /* statistic_tests.hpp header file |
---|
| 2 | * |
---|
| 3 | * Copyright Jens Maurer 2000 |
---|
| 4 | * Distributed under the Boost Software License, Version 1.0. (See |
---|
| 5 | * accompanying file LICENSE_1_0.txt or copy at |
---|
| 6 | * http://www.boost.org/LICENSE_1_0.txt) |
---|
| 7 | * |
---|
| 8 | * $Id: statistic_tests.hpp,v 1.13 2004/07/27 03:43:34 dgregor Exp $ |
---|
| 9 | * |
---|
| 10 | */ |
---|
| 11 | |
---|
| 12 | #ifndef STATISTIC_TESTS_HPP |
---|
| 13 | #define STATISTIC_TESTS_HPP |
---|
| 14 | |
---|
| 15 | #include <stdexcept> |
---|
| 16 | #include <iterator> |
---|
| 17 | #include <vector> |
---|
| 18 | #include <boost/limits.hpp> |
---|
| 19 | #include <algorithm> |
---|
| 20 | #include <cmath> |
---|
| 21 | |
---|
| 22 | #include <boost/random.hpp> |
---|
| 23 | #include <boost/config.hpp> |
---|
| 24 | |
---|
| 25 | |
---|
| 26 | #if defined(BOOST_MSVC) && BOOST_MSVC <= 1300 |
---|
| 27 | namespace std |
---|
| 28 | { |
---|
| 29 | inline double pow(double a, double b) { return ::pow(a,b); } |
---|
| 30 | inline double ceil(double x) { return ::ceil(x); } |
---|
| 31 | } // namespace std |
---|
| 32 | #endif |
---|
| 33 | |
---|
| 34 | |
---|
| 35 | template<class T> |
---|
| 36 | inline T fac(int k) |
---|
| 37 | { |
---|
| 38 | T result = 1; |
---|
| 39 | for(T i = 2; i <= k; ++i) |
---|
| 40 | result *= i; |
---|
| 41 | return result; |
---|
| 42 | } |
---|
| 43 | |
---|
| 44 | template<class T> |
---|
| 45 | T binomial(int n, int k) |
---|
| 46 | { |
---|
| 47 | if(k < n/2) |
---|
| 48 | k = n-k; |
---|
| 49 | T result = 1; |
---|
| 50 | for(int i = k+1; i<= n; ++i) |
---|
| 51 | result *= i; |
---|
| 52 | return result / fac<T>(n-k); |
---|
| 53 | } |
---|
| 54 | |
---|
| 55 | template<class T> |
---|
| 56 | T stirling2(int n, int m) |
---|
| 57 | { |
---|
| 58 | T sum = 0; |
---|
| 59 | for(int k = 0; k <= m; ++k) |
---|
| 60 | sum += binomial<T>(m, k) * std::pow(double(k), n) * |
---|
| 61 | ( (m-k)%2 == 0 ? 1 : -1); |
---|
| 62 | return sum / fac<T>(m); |
---|
| 63 | } |
---|
| 64 | |
---|
| 65 | /* |
---|
| 66 | * Experiments which create an empirical distribution in classes, |
---|
| 67 | * suitable for the chi-square test. |
---|
| 68 | */ |
---|
| 69 | // std::floor(gen() * classes) |
---|
| 70 | |
---|
| 71 | class experiment_base |
---|
| 72 | { |
---|
| 73 | public: |
---|
| 74 | experiment_base(int cls) : _classes(cls) { } |
---|
| 75 | unsigned int classes() const { return _classes; } |
---|
| 76 | protected: |
---|
| 77 | unsigned int _classes; |
---|
| 78 | }; |
---|
| 79 | |
---|
| 80 | class equidistribution_experiment : public experiment_base |
---|
| 81 | { |
---|
| 82 | public: |
---|
| 83 | explicit equidistribution_experiment(unsigned int classes) |
---|
| 84 | : experiment_base(classes) { } |
---|
| 85 | |
---|
| 86 | template<class NumberGenerator, class Counter> |
---|
| 87 | void run(NumberGenerator f, Counter & count, int n) const |
---|
| 88 | { |
---|
| 89 | assert((f.min)() == 0 && |
---|
| 90 | static_cast<unsigned int>((f.max)()) == classes()-1); |
---|
| 91 | for(int i = 0; i < n; ++i) |
---|
| 92 | count(f()); |
---|
| 93 | } |
---|
| 94 | double probability(int i) const { return 1.0/classes(); } |
---|
| 95 | }; |
---|
| 96 | |
---|
| 97 | // two-dimensional equidistribution experiment |
---|
| 98 | class equidistribution_2d_experiment : public equidistribution_experiment |
---|
| 99 | { |
---|
| 100 | public: |
---|
| 101 | explicit equidistribution_2d_experiment(unsigned int classes) |
---|
| 102 | : equidistribution_experiment(classes) { } |
---|
| 103 | |
---|
| 104 | template<class NumberGenerator, class Counter> |
---|
| 105 | void run(NumberGenerator f, Counter & count, int n) const |
---|
| 106 | { |
---|
| 107 | unsigned int range = (f.max)()+1; |
---|
| 108 | assert((f.min)() == 0 && range*range == classes()); |
---|
| 109 | for(int i = 0; i < n; ++i) { |
---|
| 110 | int y1 = f(); |
---|
| 111 | int y2 = f(); |
---|
| 112 | count(y1 + range * y2); |
---|
| 113 | } |
---|
| 114 | } |
---|
| 115 | }; |
---|
| 116 | |
---|
| 117 | // distribution experiment: assume a probability density and |
---|
| 118 | // count events so that an equidistribution results. |
---|
| 119 | class distribution_experiment : public equidistribution_experiment |
---|
| 120 | { |
---|
| 121 | public: |
---|
| 122 | template<class UnaryFunction> |
---|
| 123 | distribution_experiment(UnaryFunction probability , unsigned int classes) |
---|
| 124 | : equidistribution_experiment(classes), limit(classes) |
---|
| 125 | { |
---|
| 126 | for(unsigned int i = 0; i < classes-1; ++i) |
---|
| 127 | limit[i] = invert_monotone_inc(probability, (i+1)*0.05, 0, 1000); |
---|
| 128 | limit[classes-1] = std::numeric_limits<double>::infinity(); |
---|
| 129 | if(limit[classes-1] < (std::numeric_limits<double>::max)()) |
---|
| 130 | limit[classes-1] = (std::numeric_limits<double>::max)(); |
---|
| 131 | #if 0 |
---|
| 132 | std::cout << __PRETTY_FUNCTION__ << ": "; |
---|
| 133 | for(unsigned int i = 0; i < classes; ++i) |
---|
| 134 | std::cout << limit[i] << " "; |
---|
| 135 | std::cout << std::endl; |
---|
| 136 | #endif |
---|
| 137 | } |
---|
| 138 | |
---|
| 139 | template<class NumberGenerator, class Counter> |
---|
| 140 | void run(NumberGenerator f, Counter & count, int n) const |
---|
| 141 | { |
---|
| 142 | for(int i = 0; i < n; ++i) { |
---|
| 143 | limits_type::const_iterator it = |
---|
| 144 | std::lower_bound(limit.begin(), limit.end(), f()); |
---|
| 145 | count(it-limit.begin()); |
---|
| 146 | } |
---|
| 147 | } |
---|
| 148 | private: |
---|
| 149 | typedef std::vector<double> limits_type; |
---|
| 150 | limits_type limit; |
---|
| 151 | }; |
---|
| 152 | |
---|
| 153 | // runs-up/runs-down experiment |
---|
| 154 | template<bool up> |
---|
| 155 | class runs_experiment : public experiment_base |
---|
| 156 | { |
---|
| 157 | public: |
---|
| 158 | explicit runs_experiment(unsigned int classes) : experiment_base(classes) { } |
---|
| 159 | |
---|
| 160 | template<class UniformRandomNumberGenerator, class Counter> |
---|
| 161 | void run(UniformRandomNumberGenerator f, Counter & count, int n) const |
---|
| 162 | { |
---|
| 163 | typedef typename UniformRandomNumberGenerator::result_type result_type; |
---|
| 164 | result_type init = (up ? (f.min)() : (f.max)()); |
---|
| 165 | result_type previous = init; |
---|
| 166 | unsigned int length = 0; |
---|
| 167 | for(int i = 0; i < n; ++i) { |
---|
| 168 | result_type val = f(); |
---|
| 169 | if(up ? previous <= val : previous >= val) { |
---|
| 170 | previous = val; |
---|
| 171 | ++length; |
---|
| 172 | } else { |
---|
| 173 | count((std::min)(length, classes())-1); |
---|
| 174 | length = 0; |
---|
| 175 | previous = init; |
---|
| 176 | // don't use this value, so that runs are independent |
---|
| 177 | } |
---|
| 178 | } |
---|
| 179 | } |
---|
| 180 | double probability(unsigned int r) const |
---|
| 181 | { |
---|
| 182 | if(r == classes()-1) |
---|
| 183 | return 1.0/fac<double>(classes()); |
---|
| 184 | else |
---|
| 185 | return static_cast<double>(r+1)/fac<double>(r+2); |
---|
| 186 | } |
---|
| 187 | }; |
---|
| 188 | |
---|
| 189 | // gap length experiment |
---|
| 190 | class gap_experiment : public experiment_base |
---|
| 191 | { |
---|
| 192 | public: |
---|
| 193 | gap_experiment(unsigned int classes, double alpha, double beta) |
---|
| 194 | : experiment_base(classes), alpha(alpha), beta(beta) { } |
---|
| 195 | |
---|
| 196 | template<class UniformRandomNumberGenerator, class Counter> |
---|
| 197 | void run(UniformRandomNumberGenerator f, Counter & count, int n) const |
---|
| 198 | { |
---|
| 199 | typedef typename UniformRandomNumberGenerator::result_type result_type; |
---|
| 200 | double range = (f.max)() - (f.min)() + 1.0; |
---|
| 201 | result_type low = static_cast<result_type>(alpha * range); |
---|
| 202 | result_type high = static_cast<result_type>(beta * range); |
---|
| 203 | unsigned int length = 0; |
---|
| 204 | for(int i = 0; i < n; ) { |
---|
| 205 | result_type value = f() - (f.min)(); |
---|
| 206 | if(value < low || value > high) |
---|
| 207 | ++length; |
---|
| 208 | else { |
---|
| 209 | count((std::min)(length, classes()-1)); |
---|
| 210 | length = 0; |
---|
| 211 | ++i; |
---|
| 212 | } |
---|
| 213 | } |
---|
| 214 | } |
---|
| 215 | double probability(unsigned int r) const |
---|
| 216 | { |
---|
| 217 | double p = beta-alpha; |
---|
| 218 | if(r == classes()-1) |
---|
| 219 | return std::pow(1-p, static_cast<double>(r)); |
---|
| 220 | else |
---|
| 221 | return p * std::pow(1-p, static_cast<double>(r)); |
---|
| 222 | } |
---|
| 223 | private: |
---|
| 224 | double alpha, beta; |
---|
| 225 | }; |
---|
| 226 | |
---|
| 227 | // poker experiment |
---|
| 228 | class poker_experiment : public experiment_base |
---|
| 229 | { |
---|
| 230 | public: |
---|
| 231 | poker_experiment(unsigned int d, unsigned int k) |
---|
| 232 | : experiment_base(k), range(d) |
---|
| 233 | { |
---|
| 234 | assert(range > 1); |
---|
| 235 | } |
---|
| 236 | |
---|
| 237 | template<class UniformRandomNumberGenerator, class Counter> |
---|
| 238 | void run(UniformRandomNumberGenerator f, Counter & count, int n) const |
---|
| 239 | { |
---|
| 240 | typedef typename UniformRandomNumberGenerator::result_type result_type; |
---|
| 241 | assert(std::numeric_limits<result_type>::is_integer); |
---|
| 242 | assert((f.min)() == 0); |
---|
| 243 | assert((f.max)() == static_cast<result_type>(range-1)); |
---|
| 244 | std::vector<result_type> v(classes()); |
---|
| 245 | for(int i = 0; i < n; ++i) { |
---|
| 246 | for(unsigned int j = 0; j < classes(); ++j) |
---|
| 247 | v[j] = f(); |
---|
| 248 | std::sort(v.begin(), v.end()); |
---|
| 249 | result_type prev = v[0]; |
---|
| 250 | int r = 1; // count different values in v |
---|
| 251 | for(unsigned int i = 1; i < classes(); ++i) { |
---|
| 252 | if(prev != v[i]) { |
---|
| 253 | prev = v[i]; |
---|
| 254 | ++r; |
---|
| 255 | } |
---|
| 256 | } |
---|
| 257 | count(r-1); |
---|
| 258 | } |
---|
| 259 | } |
---|
| 260 | |
---|
| 261 | double probability(unsigned int r) const |
---|
| 262 | { |
---|
| 263 | ++r; // transform to 1 <= r <= 5 |
---|
| 264 | double result = range; |
---|
| 265 | for(unsigned int i = 1; i < r; ++i) |
---|
| 266 | result *= range-i; |
---|
| 267 | return result / std::pow(range, static_cast<double>(classes())) * |
---|
| 268 | stirling2<double>(classes(), r); |
---|
| 269 | } |
---|
| 270 | private: |
---|
| 271 | unsigned int range; |
---|
| 272 | }; |
---|
| 273 | |
---|
| 274 | // coupon collector experiment |
---|
| 275 | class coupon_collector_experiment : public experiment_base |
---|
| 276 | { |
---|
| 277 | public: |
---|
| 278 | coupon_collector_experiment(unsigned int d, unsigned int cls) |
---|
| 279 | : experiment_base(cls), d(d) |
---|
| 280 | { |
---|
| 281 | assert(d > 1); |
---|
| 282 | } |
---|
| 283 | |
---|
| 284 | template<class UniformRandomNumberGenerator, class Counter> |
---|
| 285 | void run(UniformRandomNumberGenerator f, Counter & count, int n) const |
---|
| 286 | { |
---|
| 287 | typedef typename UniformRandomNumberGenerator::result_type result_type; |
---|
| 288 | assert(std::numeric_limits<result_type>::is_integer); |
---|
| 289 | assert((f.min)() == 0); |
---|
| 290 | assert((f.max)() == static_cast<result_type>(d-1)); |
---|
| 291 | std::vector<bool> occurs(d); |
---|
| 292 | for(int i = 0; i < n; ++i) { |
---|
| 293 | occurs.assign(d, false); |
---|
| 294 | unsigned int r = 0; // length of current sequence |
---|
| 295 | int q = 0; // number of non-duplicates in current set |
---|
| 296 | for(;;) { |
---|
| 297 | result_type val = f(); |
---|
| 298 | ++r; |
---|
| 299 | if(!occurs[val]) { // new set element |
---|
| 300 | occurs[val] = true; |
---|
| 301 | ++q; |
---|
| 302 | if(q == d) |
---|
| 303 | break; // one complete set |
---|
| 304 | } |
---|
| 305 | } |
---|
| 306 | count((std::min)(r-d, classes()-1)); |
---|
| 307 | } |
---|
| 308 | } |
---|
| 309 | double probability(unsigned int r) const |
---|
| 310 | { |
---|
| 311 | if(r == classes()-1) |
---|
| 312 | return 1-fac<double>(d)/std::pow(d, static_cast<double>(d+classes()-2))* |
---|
| 313 | stirling2<double>(d+classes()-2, d); |
---|
| 314 | else |
---|
| 315 | return fac<double>(d)/std::pow(d, static_cast<double>(d+r)) * |
---|
| 316 | stirling2<double>(d+r-1, d-1); |
---|
| 317 | } |
---|
| 318 | private: |
---|
| 319 | int d; |
---|
| 320 | }; |
---|
| 321 | |
---|
| 322 | // permutation test |
---|
| 323 | class permutation_experiment : public equidistribution_experiment |
---|
| 324 | { |
---|
| 325 | public: |
---|
| 326 | permutation_experiment(unsigned int t) |
---|
| 327 | : equidistribution_experiment(fac<int>(t)), t(t) |
---|
| 328 | { |
---|
| 329 | assert(t > 1); |
---|
| 330 | } |
---|
| 331 | |
---|
| 332 | template<class UniformRandomNumberGenerator, class Counter> |
---|
| 333 | void run(UniformRandomNumberGenerator f, Counter & count, int n) const |
---|
| 334 | { |
---|
| 335 | typedef typename UniformRandomNumberGenerator::result_type result_type; |
---|
| 336 | std::vector<result_type> v(t); |
---|
| 337 | for(int i = 0; i < n; ++i) { |
---|
| 338 | std::generate_n(v.begin(), t, f); |
---|
| 339 | int x = 0; |
---|
| 340 | for(int r = t-1; r > 0; r--) { |
---|
| 341 | typename std::vector<result_type>::iterator it = |
---|
| 342 | std::max_element(v.begin(), v.begin()+r+1); |
---|
| 343 | x = (r+1)*x + (it-v.begin()); |
---|
| 344 | std::iter_swap(it, v.begin()+r); |
---|
| 345 | } |
---|
| 346 | count(x); |
---|
| 347 | } |
---|
| 348 | } |
---|
| 349 | private: |
---|
| 350 | int t; |
---|
| 351 | }; |
---|
| 352 | |
---|
| 353 | // birthday spacing experiment test |
---|
| 354 | class birthday_spacing_experiment : public experiment_base |
---|
| 355 | { |
---|
| 356 | public: |
---|
| 357 | birthday_spacing_experiment(unsigned int d, int n, int m) |
---|
| 358 | : experiment_base(d), n(n), m(m) |
---|
| 359 | { |
---|
| 360 | } |
---|
| 361 | |
---|
| 362 | template<class UniformRandomNumberGenerator, class Counter> |
---|
| 363 | void run(UniformRandomNumberGenerator f, Counter & count, int n_total) const |
---|
| 364 | { |
---|
| 365 | typedef typename UniformRandomNumberGenerator::result_type result_type; |
---|
| 366 | assert(std::numeric_limits<result_type>::is_integer); |
---|
| 367 | assert((f.min)() == 0); |
---|
| 368 | assert((f.max)() == static_cast<result_type>(m-1)); |
---|
| 369 | |
---|
| 370 | for(int j = 0; j < n_total; j++) { |
---|
| 371 | std::vector<result_type> v(n); |
---|
| 372 | std::generate_n(v.begin(), n, f); |
---|
| 373 | std::sort(v.begin(), v.end()); |
---|
| 374 | std::vector<result_type> spacing(n); |
---|
| 375 | for(int i = 0; i < n-1; i++) |
---|
| 376 | spacing[i] = v[i+1]-v[i]; |
---|
| 377 | spacing[n-1] = v[0] + m - v[n-1]; |
---|
| 378 | std::sort(spacing.begin(), spacing.end()); |
---|
| 379 | unsigned int k = 0; |
---|
| 380 | for(int i = 0; i < n-1; ++i) { |
---|
| 381 | if(spacing[i] == spacing[i+1]) |
---|
| 382 | ++k; |
---|
| 383 | } |
---|
| 384 | count((std::min)(k, classes()-1)); |
---|
| 385 | } |
---|
| 386 | } |
---|
| 387 | |
---|
| 388 | double probability(unsigned int r) const |
---|
| 389 | { |
---|
| 390 | assert(classes() == 4); |
---|
| 391 | assert(m == (1<<25)); |
---|
| 392 | assert(n == 512); |
---|
| 393 | static const double prob[] = { 0.368801577, 0.369035243, 0.183471182, |
---|
| 394 | 0.078691997 }; |
---|
| 395 | return prob[r]; |
---|
| 396 | } |
---|
| 397 | private: |
---|
| 398 | int n, m; |
---|
| 399 | }; |
---|
| 400 | /* |
---|
| 401 | * Misc. helper functions. |
---|
| 402 | */ |
---|
| 403 | |
---|
| 404 | template<class Float> |
---|
| 405 | struct distribution_function |
---|
| 406 | { |
---|
| 407 | typedef Float result_type; |
---|
| 408 | typedef Float argument_type; |
---|
| 409 | typedef Float first_argument_type; |
---|
| 410 | typedef Float second_argument_type; |
---|
| 411 | }; |
---|
| 412 | |
---|
| 413 | // computes P(K_n <= t) or P(t1 <= K_n <= t2). See Knuth, 3.3.1 |
---|
| 414 | class kolmogorov_smirnov_probability : public distribution_function<double> |
---|
| 415 | { |
---|
| 416 | public: |
---|
| 417 | kolmogorov_smirnov_probability(int n) |
---|
| 418 | : approx(n > 50), n(n), sqrt_n(std::sqrt(double(n))) |
---|
| 419 | { |
---|
| 420 | if(!approx) |
---|
| 421 | n_n = std::pow(static_cast<double>(n), n); |
---|
| 422 | } |
---|
| 423 | |
---|
| 424 | double operator()(double t) const |
---|
| 425 | { |
---|
| 426 | if(approx) { |
---|
| 427 | return 1-std::exp(-2*t*t)*(1-2.0/3.0*t/sqrt_n); |
---|
| 428 | } else { |
---|
| 429 | t *= sqrt_n; |
---|
| 430 | double sum = 0; |
---|
| 431 | for(int k = static_cast<int>(std::ceil(t)); k <= n; k++) |
---|
| 432 | sum += binomial<double>(n, k) * std::pow(k-t, k) * |
---|
| 433 | std::pow(t+n-k, n-k-1); |
---|
| 434 | return 1 - t/n_n * sum; |
---|
| 435 | } |
---|
| 436 | } |
---|
| 437 | double operator()(double t1, double t2) const |
---|
| 438 | { return operator()(t2) - operator()(t1); } |
---|
| 439 | |
---|
| 440 | private: |
---|
| 441 | bool approx; |
---|
| 442 | int n; |
---|
| 443 | double sqrt_n; |
---|
| 444 | double n_n; |
---|
| 445 | }; |
---|
| 446 | |
---|
| 447 | /* |
---|
| 448 | * Experiments for generators with continuous distribution functions |
---|
| 449 | */ |
---|
| 450 | class kolmogorov_experiment |
---|
| 451 | { |
---|
| 452 | public: |
---|
| 453 | kolmogorov_experiment(int n) : n(n), ksp(n) { } |
---|
| 454 | template<class NumberGenerator, class Distribution> |
---|
| 455 | double run(NumberGenerator gen, Distribution distrib) const |
---|
| 456 | { |
---|
| 457 | const int m = n; |
---|
| 458 | typedef std::vector<double> saved_temp; |
---|
| 459 | saved_temp a(m,1.0), b(m,0); |
---|
| 460 | std::vector<int> c(m,0); |
---|
| 461 | for(int i = 0; i < n; ++i) { |
---|
| 462 | double val = gen(); |
---|
| 463 | double y = distrib(val); |
---|
| 464 | int k = static_cast<int>(std::floor(m*y)); |
---|
| 465 | if(k >= m) |
---|
| 466 | --k; // should not happen |
---|
| 467 | a[k] = (std::min)(a[k], y); |
---|
| 468 | b[k] = (std::max)(b[k], y); |
---|
| 469 | ++c[k]; |
---|
| 470 | } |
---|
| 471 | double kplus = 0, kminus = 0; |
---|
| 472 | int j = 0; |
---|
| 473 | for(int k = 0; k < m; ++k) { |
---|
| 474 | if(c[k] > 0) { |
---|
| 475 | kminus = (std::max)(kminus, a[k]-j/static_cast<double>(n)); |
---|
| 476 | j += c[k]; |
---|
| 477 | kplus = (std::max)(kplus, j/static_cast<double>(n) - b[k]); |
---|
| 478 | } |
---|
| 479 | } |
---|
| 480 | kplus *= std::sqrt(double(n)); |
---|
| 481 | kminus *= std::sqrt(double(n)); |
---|
| 482 | // std::cout << "k+ " << kplus << " k- " << kminus << std::endl; |
---|
| 483 | return kplus; |
---|
| 484 | } |
---|
| 485 | double probability(double x) const |
---|
| 486 | { |
---|
| 487 | return ksp(x); |
---|
| 488 | } |
---|
| 489 | private: |
---|
| 490 | int n; |
---|
| 491 | kolmogorov_smirnov_probability ksp; |
---|
| 492 | }; |
---|
| 493 | |
---|
| 494 | // maximum-of-t test (KS-based) |
---|
| 495 | template<class UniformRandomNumberGenerator> |
---|
| 496 | class maximum_experiment |
---|
| 497 | { |
---|
| 498 | public: |
---|
| 499 | typedef UniformRandomNumberGenerator base_type; |
---|
| 500 | maximum_experiment(base_type & f, int n, int t) : f(f), ke(n), t(t) |
---|
| 501 | { } |
---|
| 502 | |
---|
| 503 | double operator()() const |
---|
| 504 | { |
---|
| 505 | double res = ke.run(generator(f, t), |
---|
| 506 | std::bind2nd(std::ptr_fun(static_cast<double (*)(double, double)>(&std::pow)), t)); |
---|
| 507 | return res; |
---|
| 508 | } |
---|
| 509 | |
---|
| 510 | private: |
---|
| 511 | struct generator { |
---|
| 512 | generator(base_type & f, int t) : f(f), t(t) { } |
---|
| 513 | double operator()() |
---|
| 514 | { |
---|
| 515 | double mx = f(); |
---|
| 516 | for(int i = 1; i < t; ++i) |
---|
| 517 | mx = (std::max)(mx, f()); |
---|
| 518 | return mx; |
---|
| 519 | } |
---|
| 520 | private: |
---|
| 521 | boost::uniform_01<base_type> f; |
---|
| 522 | int t; |
---|
| 523 | }; |
---|
| 524 | base_type & f; |
---|
| 525 | kolmogorov_experiment ke; |
---|
| 526 | int t; |
---|
| 527 | }; |
---|
| 528 | |
---|
| 529 | // compute a chi-square value for the distribution approximation error |
---|
| 530 | template<class ForwardIterator, class UnaryFunction> |
---|
| 531 | typename UnaryFunction::result_type |
---|
| 532 | chi_square_value(ForwardIterator first, ForwardIterator last, |
---|
| 533 | UnaryFunction probability) |
---|
| 534 | { |
---|
| 535 | typedef std::iterator_traits<ForwardIterator> iter_traits; |
---|
| 536 | typedef typename iter_traits::value_type counter_type; |
---|
| 537 | typedef typename UnaryFunction::result_type result_type; |
---|
| 538 | unsigned int classes = std::distance(first, last); |
---|
| 539 | result_type sum = 0; |
---|
| 540 | counter_type n = 0; |
---|
| 541 | for(unsigned int i = 0; i < classes; ++first, ++i) { |
---|
| 542 | counter_type count = *first; |
---|
| 543 | n += count; |
---|
| 544 | sum += (count/probability(i)) * count; // avoid overflow |
---|
| 545 | } |
---|
| 546 | #if 0 |
---|
| 547 | for(unsigned int i = 0; i < classes; ++i) { |
---|
| 548 | // std::cout << (n*probability(i)) << " "; |
---|
| 549 | if(n * probability(i) < 5) |
---|
| 550 | std::cerr << "Not enough test runs for slot " << i |
---|
| 551 | << " p=" << probability(i) << ", n=" << n |
---|
| 552 | << std::endl; |
---|
| 553 | } |
---|
| 554 | #endif |
---|
| 555 | // std::cout << std::endl; |
---|
| 556 | // throw std::invalid_argument("not enough test runs"); |
---|
| 557 | |
---|
| 558 | return sum/n - n; |
---|
| 559 | } |
---|
| 560 | template<class RandomAccessContainer> |
---|
| 561 | class generic_counter |
---|
| 562 | { |
---|
| 563 | public: |
---|
| 564 | explicit generic_counter(unsigned int classes) : container(classes, 0) { } |
---|
| 565 | void operator()(int i) |
---|
| 566 | { |
---|
| 567 | assert(i >= 0); |
---|
| 568 | assert(static_cast<unsigned int>(i) < container.size()); |
---|
| 569 | ++container[i]; |
---|
| 570 | } |
---|
| 571 | typename RandomAccessContainer::const_iterator begin() const |
---|
| 572 | { return container.begin(); } |
---|
| 573 | typename RandomAccessContainer::const_iterator end() const |
---|
| 574 | { return container.end(); } |
---|
| 575 | |
---|
| 576 | private: |
---|
| 577 | RandomAccessContainer container; |
---|
| 578 | }; |
---|
| 579 | |
---|
| 580 | // chi_square test |
---|
| 581 | template<class Experiment, class Generator> |
---|
| 582 | double run_experiment(const Experiment & experiment, Generator gen, int n) |
---|
| 583 | { |
---|
| 584 | generic_counter<std::vector<int> > v(experiment.classes()); |
---|
| 585 | experiment.run(gen, v, n); |
---|
| 586 | return chi_square_value(v.begin(), v.end(), |
---|
| 587 | std::bind1st(std::mem_fun_ref(&Experiment::probability), |
---|
| 588 | experiment)); |
---|
| 589 | } |
---|
| 590 | |
---|
| 591 | // number generator with experiment results (for nesting) |
---|
| 592 | template<class Experiment, class Generator> |
---|
| 593 | class experiment_generator_t |
---|
| 594 | { |
---|
| 595 | public: |
---|
| 596 | experiment_generator_t(const Experiment & exper, Generator & gen, int n) |
---|
| 597 | : experiment(exper), generator(gen), n(n) { } |
---|
| 598 | double operator()() { return run_experiment(experiment, generator, n); } |
---|
| 599 | private: |
---|
| 600 | const Experiment & experiment; |
---|
| 601 | Generator & generator; |
---|
| 602 | int n; |
---|
| 603 | }; |
---|
| 604 | |
---|
| 605 | template<class Experiment, class Generator> |
---|
| 606 | experiment_generator_t<Experiment, Generator> |
---|
| 607 | experiment_generator(const Experiment & e, Generator & gen, int n) |
---|
| 608 | { |
---|
| 609 | return experiment_generator_t<Experiment, Generator>(e, gen, n); |
---|
| 610 | } |
---|
| 611 | |
---|
| 612 | |
---|
| 613 | template<class Experiment, class Generator, class Distribution> |
---|
| 614 | class ks_experiment_generator_t |
---|
| 615 | { |
---|
| 616 | public: |
---|
| 617 | ks_experiment_generator_t(const Experiment & exper, Generator & gen, |
---|
| 618 | const Distribution & distrib) |
---|
| 619 | : experiment(exper), generator(gen), distribution(distrib) { } |
---|
| 620 | double operator()() { return experiment.run(generator, distribution); } |
---|
| 621 | private: |
---|
| 622 | const Experiment & experiment; |
---|
| 623 | Generator & generator; |
---|
| 624 | Distribution distribution; |
---|
| 625 | }; |
---|
| 626 | |
---|
| 627 | template<class Experiment, class Generator, class Distribution> |
---|
| 628 | ks_experiment_generator_t<Experiment, Generator, Distribution> |
---|
| 629 | ks_experiment_generator(const Experiment & e, Generator & gen, |
---|
| 630 | const Distribution & distrib) |
---|
| 631 | { |
---|
| 632 | return ks_experiment_generator_t<Experiment, Generator, Distribution> |
---|
| 633 | (e, gen, distrib); |
---|
| 634 | } |
---|
| 635 | |
---|
| 636 | |
---|
| 637 | #endif /* STATISTIC_TESTS_HPP */ |
---|
| 638 | |
---|