| 1 |
// boost\math\distributions\non_central_beta.hpp |
| 2 |
|
| 3 |
// Copyright John Maddock 2008. |
| 4 |
// Copyright Matt Borland 2024. |
| 5 |
// Use, modification and distribution are subject to the |
| 6 |
// Boost Software License, Version 1.0. |
| 7 |
// (See accompanying file LICENSE_1_0.txt |
| 8 |
// or copy at http://www.boost.org/LICENSE_1_0.txt) |
| 9 |
|
| 10 |
#ifndef BOOST_MATH_SPECIAL_NON_CENTRAL_BETA_HPP |
| 11 |
#define BOOST_MATH_SPECIAL_NON_CENTRAL_BETA_HPP |
| 12 |
|
| 13 |
#include <boost/math/tools/config.hpp> |
| 14 |
#include <boost/math/tools/tuple.hpp> |
| 15 |
#include <boost/math/tools/cstdint.hpp> |
| 16 |
#include <boost/math/tools/numeric_limits.hpp> |
| 17 |
#include <boost/math/distributions/fwd.hpp> |
| 18 |
#include <boost/math/special_functions/beta.hpp> // for incomplete gamma. gamma_q |
| 19 |
#include <boost/math/distributions/complement.hpp> // complements |
| 20 |
#include <boost/math/distributions/beta.hpp> // central distribution |
| 21 |
#include <boost/math/distributions/detail/generic_mode.hpp> |
| 22 |
#include <boost/math/distributions/detail/common_error_handling.hpp> // error checks |
| 23 |
#include <boost/math/special_functions/fpclassify.hpp> // isnan. |
| 24 |
#include <boost/math/special_functions/trunc.hpp> |
| 25 |
#include <boost/math/tools/roots.hpp> // for root finding. |
| 26 |
#include <boost/math/tools/series.hpp> |
| 27 |
#include <boost/math/policies/error_handling.hpp> |
| 28 |
|
| 29 |
namespace boost |
| 30 |
{ |
| 31 |
namespace math |
| 32 |
{ |
| 33 |
|
| 34 |
template <class RealType, class Policy> |
| 35 |
class non_central_beta_distribution; |
| 36 |
|
| 37 |
namespace detail{ |
| 38 |
|
| 39 |
template <class T, class Policy> |
| 40 |
BOOST_MATH_GPU_ENABLED T non_central_beta_p(T a, T b, T lam, T x, T y, const Policy& pol, T init_val = 0) |
| 41 |
{ |
| 42 |
BOOST_MATH_STD_USING |
| 43 |
using namespace boost::math; |
| 44 |
// |
| 45 |
// Variables come first: |
| 46 |
// |
| 47 |
boost::math::uintmax_t max_iter = policies::get_max_series_iterations<Policy>(); |
| 48 |
T errtol = boost::math::policies::get_epsilon<T, Policy>(); |
| 49 |
T l2 = lam / 2; |
| 50 |
// |
| 51 |
// k is the starting point for iteration, and is the |
| 52 |
// maximum of the poisson weighting term, |
| 53 |
// note that unlike other similar code, we do not set |
| 54 |
// k to zero, when l2 is small, as forward iteration |
| 55 |
// is unstable: |
| 56 |
// |
| 57 |
long long k = lltrunc(l2); |
| 58 |
if(k == 0) |
| 59 |
k = 1; |
| 60 |
// Starting Poisson weight: |
| 61 |
T pois = gamma_p_derivative(T(k+1), l2, pol); |
| 62 |
if(pois == 0) |
| 63 |
return init_val; |
| 64 |
// recurance term: |
| 65 |
T xterm; |
| 66 |
// Starting beta term: |
| 67 |
T beta = x < y |
| 68 |
? detail::ibeta_imp(T(a + k), b, x, pol, false, true, &xterm) |
| 69 |
: detail::ibeta_imp(b, T(a + k), y, pol, true, true, &xterm); |
| 70 |
|
| 71 |
while (fabs(beta * pois) < tools::min_value<T>()) |
| 72 |
{ |
| 73 |
if ((k == 0) || (pois == 0)) |
| 74 |
return init_val; |
| 75 |
k /= 2; |
| 76 |
pois = gamma_p_derivative(T(k + 1), l2, pol); |
| 77 |
beta = x < y |
| 78 |
? detail::ibeta_imp(T(a + k), b, x, pol, false, true, &xterm) |
| 79 |
: detail::ibeta_imp(b, T(a + k), y, pol, true, true, &xterm); |
| 80 |
} |
| 81 |
|
| 82 |
xterm *= y / (a + b + k - 1); |
| 83 |
T poisf(pois), betaf(beta), xtermf(xterm); |
| 84 |
T sum = init_val; |
| 85 |
|
| 86 |
if((beta == 0) && (xterm == 0)) |
| 87 |
return init_val; |
| 88 |
|
| 89 |
// |
| 90 |
// Backwards recursion first, this is the stable |
| 91 |
// direction for recursion: |
| 92 |
// |
| 93 |
T last_term = 0; |
| 94 |
boost::math::uintmax_t count = k; |
| 95 |
for(auto i = k; i >= 0; --i) |
| 96 |
{ |
| 97 |
T term = beta * pois; |
| 98 |
sum += term; |
| 99 |
if(((fabs(term/sum) < errtol) && (fabs(last_term) >= fabs(term))) || (term == 0)) |
| 100 |
{ |
| 101 |
count = k - i; |
| 102 |
break; |
| 103 |
} |
| 104 |
pois *= i / l2; |
| 105 |
beta += xterm; |
| 106 |
|
| 107 |
if (a + b + i != 2) |
| 108 |
{ |
| 109 |
xterm *= (a + i - 1) / (x * (a + b + i - 2)); |
| 110 |
} |
| 111 |
|
| 112 |
last_term = term; |
| 113 |
} |
| 114 |
last_term = 0; |
| 115 |
for(auto i = k + 1; ; ++i) |
| 116 |
{ |
| 117 |
poisf *= l2 / i; |
| 118 |
xtermf *= (x * (a + b + i - 2)) / (a + i - 1); |
| 119 |
betaf -= xtermf; |
| 120 |
|
| 121 |
T term = poisf * betaf; |
| 122 |
sum += term; |
| 123 |
if(((fabs(term/sum) < errtol) && (fabs(last_term) >= fabs(term))) || (term == 0)) |
| 124 |
{ |
| 125 |
break; |
| 126 |
} |
| 127 |
last_term = term; |
| 128 |
if(static_cast<boost::math::uintmax_t>(count + i - k) > max_iter) |
| 129 |
{ |
| 130 |
return policies::raise_evaluation_error("cdf(non_central_beta_distribution<%1%>, %1%)", "Series did not converge, closest value was %1%", sum, pol); // LCOV_EXCL_LINE |
| 131 |
} |
| 132 |
} |
| 133 |
return sum; |
| 134 |
} |
| 135 |
|
| 136 |
template <class T, class Policy> |
| 137 |
BOOST_MATH_GPU_ENABLED T non_central_beta_q(T a, T b, T lam, T x, T y, const Policy& pol, T init_val = 0) |
| 138 |
{ |
| 139 |
BOOST_MATH_STD_USING |
| 140 |
using namespace boost::math; |
| 141 |
// |
| 142 |
// Variables come first: |
| 143 |
// |
| 144 |
boost::math::uintmax_t max_iter = policies::get_max_series_iterations<Policy>(); |
| 145 |
T errtol = boost::math::policies::get_epsilon<T, Policy>(); |
| 146 |
T l2 = lam / 2; |
| 147 |
// |
| 148 |
// k is the starting point for iteration, and is the |
| 149 |
// maximum of the poisson weighting term: |
| 150 |
// |
| 151 |
long long k = lltrunc(l2); |
| 152 |
T pois; |
| 153 |
if(k <= 30) |
| 154 |
{ |
| 155 |
// |
| 156 |
// Might as well start at 0 since we'll likely have this number of terms anyway: |
| 157 |
// |
| 158 |
if(a + b > 1) |
| 159 |
k = 0; |
| 160 |
else if(k == 0) |
| 161 |
k = 1; |
| 162 |
} |
| 163 |
if(k == 0) |
| 164 |
{ |
| 165 |
// Starting Poisson weight: |
| 166 |
pois = exp(-l2); |
| 167 |
} |
| 168 |
else |
| 169 |
{ |
| 170 |
// Starting Poisson weight: |
| 171 |
pois = gamma_p_derivative(T(k+1), l2, pol); |
| 172 |
} |
| 173 |
if(pois == 0) |
| 174 |
return init_val; |
| 175 |
// recurance term: |
| 176 |
T xterm; |
| 177 |
// Starting beta term: |
| 178 |
T beta = x < y |
| 179 |
? detail::ibeta_imp(T(a + k), b, x, pol, true, true, &xterm) |
| 180 |
: detail::ibeta_imp(b, T(a + k), y, pol, false, true, &xterm); |
| 181 |
|
| 182 |
xterm *= y / (a + b + k - 1); |
| 183 |
T poisf(pois), betaf(beta), xtermf(xterm); |
| 184 |
T sum = init_val; |
| 185 |
if((beta == 0) && (xterm == 0)) |
| 186 |
return init_val; |
| 187 |
// |
| 188 |
// Forwards recursion first, this is the stable |
| 189 |
// direction for recursion, and the location |
| 190 |
// of the bulk of the sum: |
| 191 |
// |
| 192 |
T last_term = 0; |
| 193 |
boost::math::uintmax_t count = 0; |
| 194 |
for(long long i = k + 1; ; ++i) |
| 195 |
{ |
| 196 |
poisf *= l2 / i; |
| 197 |
xtermf *= (x * (a + b + (i - 2))) / (a + (i - 1)); |
| 198 |
betaf += xtermf; |
| 199 |
|
| 200 |
T term = poisf * betaf; |
| 201 |
sum += term; |
| 202 |
if((fabs(term/sum) < errtol) && (last_term >= term)) |
| 203 |
{ |
| 204 |
count = i - k; |
| 205 |
break; |
| 206 |
} |
| 207 |
if(static_cast<boost::math::uintmax_t>(i - k) > max_iter) |
| 208 |
{ |
| 209 |
return policies::raise_evaluation_error("cdf(non_central_beta_distribution<%1%>, %1%)", "Series did not converge, closest value was %1%", sum, pol); // LCOV_EXCL_LINE |
| 210 |
} |
| 211 |
last_term = term; |
| 212 |
} |
| 213 |
for(auto i = k; i >= 0; --i) |
| 214 |
{ |
| 215 |
T term = beta * pois; |
| 216 |
sum += term; |
| 217 |
if(fabs(term/sum) < errtol) |
| 218 |
{ |
| 219 |
break; |
| 220 |
} |
| 221 |
if(static_cast<boost::math::uintmax_t>(count + k - i) > max_iter) |
| 222 |
{ |
| 223 |
return policies::raise_evaluation_error("cdf(non_central_beta_distribution<%1%>, %1%)", "Series did not converge, closest value was %1%", sum, pol); // LCOV_EXCL_LINE |
| 224 |
} |
| 225 |
pois *= i / l2; |
| 226 |
beta -= xterm; |
| 227 |
if (a + b + i - 2 != 0) |
| 228 |
{ |
| 229 |
xterm *= (a + i - 1) / (x * (a + b + i - 2)); |
| 230 |
} |
| 231 |
} |
| 232 |
return sum; |
| 233 |
} |
| 234 |
|
| 235 |
template <class RealType, class Policy> |
| 236 |
BOOST_MATH_GPU_ENABLED inline RealType non_central_beta_cdf(RealType x, RealType y, RealType a, RealType b, RealType l, bool invert, const Policy&) |
| 237 |
{ |
| 238 |
typedef typename policies::evaluation<RealType, Policy>::type value_type; |
| 239 |
typedef typename policies::normalise< |
| 240 |
Policy, |
| 241 |
policies::promote_float<false>, |
| 242 |
policies::promote_double<false>, |
| 243 |
policies::discrete_quantile<>, |
| 244 |
policies::assert_undefined<> >::type forwarding_policy; |
| 245 |
|
| 246 |
BOOST_MATH_STD_USING |
| 247 |
|
| 248 |
if(x == 0) |
| 249 |
return invert ? 1.0f : 0.0f; |
| 250 |
if(y == 0) |
| 251 |
return invert ? 0.0f : 1.0f; |
| 252 |
value_type result; |
| 253 |
value_type c = a + b + l / 2; |
| 254 |
value_type cross = 1 - (b / c) * (1 + l / (2 * c * c)); |
| 255 |
if(l == 0) |
| 256 |
result = cdf(boost::math::beta_distribution<RealType, Policy>(a, b), x); |
| 257 |
else if(x > cross) |
| 258 |
{ |
| 259 |
// Complement is the smaller of the two: |
| 260 |
result = detail::non_central_beta_q( |
| 261 |
static_cast<value_type>(a), |
| 262 |
static_cast<value_type>(b), |
| 263 |
static_cast<value_type>(l), |
| 264 |
static_cast<value_type>(x), |
| 265 |
static_cast<value_type>(y), |
| 266 |
forwarding_policy(), |
| 267 |
static_cast<value_type>(invert ? 0 : -1)); |
| 268 |
invert = !invert; |
| 269 |
} |
| 270 |
else |
| 271 |
{ |
| 272 |
result = detail::non_central_beta_p( |
| 273 |
static_cast<value_type>(a), |
| 274 |
static_cast<value_type>(b), |
| 275 |
static_cast<value_type>(l), |
| 276 |
static_cast<value_type>(x), |
| 277 |
static_cast<value_type>(y), |
| 278 |
forwarding_policy(), |
| 279 |
static_cast<value_type>(invert ? -1 : 0)); |
| 280 |
} |
| 281 |
if(invert) |
| 282 |
result = -result; |
| 283 |
return policies::checked_narrowing_cast<RealType, forwarding_policy>( |
| 284 |
result, |
| 285 |
"boost::math::non_central_beta_cdf<%1%>(%1%, %1%, %1%)"); |
| 286 |
} |
| 287 |
|
| 288 |
template <class T, class Policy> |
| 289 |
struct nc_beta_quantile_functor |
| 290 |
{ |
| 291 |
BOOST_MATH_GPU_ENABLED nc_beta_quantile_functor(const non_central_beta_distribution<T,Policy>& d, T t, bool c) |
| 292 |
: dist(d), target(t), comp(c) {} |
| 293 |
|
| 294 |
BOOST_MATH_GPU_ENABLED T operator()(const T& x) |
| 295 |
{ |
| 296 |
return comp ? |
| 297 |
T(target - cdf(complement(dist, x))) |
| 298 |
: T(cdf(dist, x) - target); |
| 299 |
} |
| 300 |
|
| 301 |
private: |
| 302 |
non_central_beta_distribution<T,Policy> dist; |
| 303 |
T target; |
| 304 |
bool comp; |
| 305 |
}; |
| 306 |
|
| 307 |
// |
| 308 |
// This is more or less a copy of bracket_and_solve_root, but |
| 309 |
// modified to search only the interval [0,1] using similar |
| 310 |
// heuristics. |
| 311 |
// |
| 312 |
template <class F, class T, class Tol, class Policy> |
| 313 |
BOOST_MATH_GPU_ENABLED boost::math::pair<T, T> bracket_and_solve_root_01(F f, const T& guess, T factor, bool rising, Tol tol, boost::math::uintmax_t& max_iter, const Policy& pol) |
| 314 |
{ |
| 315 |
BOOST_MATH_STD_USING |
| 316 |
constexpr auto function = "boost::math::tools::bracket_and_solve_root_01<%1%>"; |
| 317 |
// |
| 318 |
// Set up initial brackets: |
| 319 |
// |
| 320 |
T a = guess; |
| 321 |
T b = a; |
| 322 |
T fa = f(a); |
| 323 |
T fb = fa; |
| 324 |
// |
| 325 |
// Set up invocation count: |
| 326 |
// |
| 327 |
boost::math::uintmax_t count = max_iter - 1; |
| 328 |
|
| 329 |
if((fa < 0) == (guess < 0 ? !rising : rising)) |
| 330 |
{ |
| 331 |
// |
| 332 |
// Zero is to the right of b, so walk upwards |
| 333 |
// until we find it: |
| 334 |
// |
| 335 |
while((boost::math::sign)(fb) == (boost::math::sign)(fa)) |
| 336 |
{ |
| 337 |
if(count == 0) |
| 338 |
{ |
| 339 |
b = policies::raise_evaluation_error(function, "Unable to bracket root, last nearest value was %1%", b, pol); // LCOV_EXCL_LINE |
| 340 |
return boost::math::make_pair(a, b); |
| 341 |
} |
| 342 |
// |
| 343 |
// Heuristic: every 20 iterations we double the growth factor in case the |
| 344 |
// initial guess was *really* bad ! |
| 345 |
// |
| 346 |
if((max_iter - count) % 20 == 0) |
| 347 |
factor *= 2; |
| 348 |
// |
| 349 |
// Now go ahead and move are guess by "factor", |
| 350 |
// we do this by reducing 1-guess by factor: |
| 351 |
// |
| 352 |
a = b; |
| 353 |
fa = fb; |
| 354 |
b = 1 - ((1 - b) / factor); |
| 355 |
fb = f(b); |
| 356 |
--count; |
| 357 |
BOOST_MATH_INSTRUMENT_CODE("a = " << a << " b = " << b << " fa = " << fa << " fb = " << fb << " count = " << count); |
| 358 |
} |
| 359 |
} |
| 360 |
else |
| 361 |
{ |
| 362 |
// |
| 363 |
// Zero is to the left of a, so walk downwards |
| 364 |
// until we find it: |
| 365 |
// |
| 366 |
while((boost::math::sign)(fb) == (boost::math::sign)(fa)) |
| 367 |
{ |
| 368 |
if(fabs(a) < tools::min_value<T>()) |
| 369 |
{ |
| 370 |
// Escape route just in case the answer is zero! |
| 371 |
max_iter -= count; |
| 372 |
max_iter += 1; |
| 373 |
return a > 0 ? boost::math::make_pair(T(0), T(a)) : boost::math::make_pair(T(a), T(0)); |
| 374 |
} |
| 375 |
if(count == 0) |
| 376 |
{ |
| 377 |
a = policies::raise_evaluation_error(function, "Unable to bracket root, last nearest value was %1%", a, pol); // LCOV_EXCL_LINE |
| 378 |
return boost::math::make_pair(a, b); |
| 379 |
} |
| 380 |
// |
| 381 |
// Heuristic: every 20 iterations we double the growth factor in case the |
| 382 |
// initial guess was *really* bad ! |
| 383 |
// |
| 384 |
if((max_iter - count) % 20 == 0) |
| 385 |
factor *= 2; |
| 386 |
// |
| 387 |
// Now go ahead and move are guess by "factor": |
| 388 |
// |
| 389 |
b = a; |
| 390 |
fb = fa; |
| 391 |
a /= factor; |
| 392 |
fa = f(a); |
| 393 |
--count; |
| 394 |
BOOST_MATH_INSTRUMENT_CODE("a = " << a << " b = " << b << " fa = " << fa << " fb = " << fb << " count = " << count); |
| 395 |
} |
| 396 |
} |
| 397 |
max_iter -= count; |
| 398 |
max_iter += 1; |
| 399 |
boost::math::pair<T, T> r = toms748_solve( |
| 400 |
f, |
| 401 |
(a < 0 ? b : a), |
| 402 |
(a < 0 ? a : b), |
| 403 |
(a < 0 ? fb : fa), |
| 404 |
(a < 0 ? fa : fb), |
| 405 |
tol, |
| 406 |
count, |
| 407 |
pol); |
| 408 |
max_iter += count; |
| 409 |
BOOST_MATH_INSTRUMENT_CODE("max_iter = " << max_iter << " count = " << count); |
| 410 |
return r; |
| 411 |
} |
| 412 |
|
| 413 |
template <class RealType, class Policy> |
| 414 |
BOOST_MATH_GPU_ENABLED RealType nc_beta_quantile(const non_central_beta_distribution<RealType, Policy>& dist, const RealType& p, bool comp) |
| 415 |
{ |
| 416 |
constexpr auto function = "quantile(non_central_beta_distribution<%1%>, %1%)"; |
| 417 |
typedef typename policies::evaluation<RealType, Policy>::type value_type; |
| 418 |
typedef typename policies::normalise< |
| 419 |
Policy, |
| 420 |
policies::promote_float<false>, |
| 421 |
policies::promote_double<false>, |
| 422 |
policies::discrete_quantile<>, |
| 423 |
policies::assert_undefined<> >::type forwarding_policy; |
| 424 |
|
| 425 |
value_type a = dist.alpha(); |
| 426 |
value_type b = dist.beta(); |
| 427 |
value_type l = dist.non_centrality(); |
| 428 |
value_type r; |
| 429 |
if(!beta_detail::check_alpha( |
| 430 |
function, |
| 431 |
a, &r, Policy()) |
| 432 |
|| |
| 433 |
!beta_detail::check_beta( |
| 434 |
function, |
| 435 |
b, &r, Policy()) |
| 436 |
|| |
| 437 |
!detail::check_non_centrality( |
| 438 |
function, |
| 439 |
l, |
| 440 |
&r, |
| 441 |
Policy()) |
| 442 |
|| |
| 443 |
!detail::check_probability( |
| 444 |
function, |
| 445 |
static_cast<value_type>(p), |
| 446 |
&r, |
| 447 |
Policy())) |
| 448 |
return static_cast<RealType>(r); |
| 449 |
// |
| 450 |
// Special cases first: |
| 451 |
// |
| 452 |
if(p == 0) |
| 453 |
return comp |
| 454 |
? 1.0f |
| 455 |
: 0.0f; |
| 456 |
if(p == 1) |
| 457 |
return !comp |
| 458 |
? 1.0f |
| 459 |
: 0.0f; |
| 460 |
|
| 461 |
value_type c = a + b + l / 2; |
| 462 |
value_type mean = 1 - (b / c) * (1 + l / (2 * c * c)); |
| 463 |
/* |
| 464 |
// |
| 465 |
// Calculate a normal approximation to the quantile, |
| 466 |
// uses mean and variance approximations from: |
| 467 |
// Algorithm AS 310: |
| 468 |
// Computing the Non-Central Beta Distribution Function |
| 469 |
// R. Chattamvelli; R. Shanmugam |
| 470 |
// Applied Statistics, Vol. 46, No. 1. (1997), pp. 146-156. |
| 471 |
// |
| 472 |
// Unfortunately, when this is wrong it tends to be *very* |
| 473 |
// wrong, so it's disabled for now, even though it often |
| 474 |
// gets the initial guess quite close. Probably we could |
| 475 |
// do much better by factoring in the skewness if only |
| 476 |
// we could calculate it.... |
| 477 |
// |
| 478 |
value_type delta = l / 2; |
| 479 |
value_type delta2 = delta * delta; |
| 480 |
value_type delta3 = delta * delta2; |
| 481 |
value_type delta4 = delta2 * delta2; |
| 482 |
value_type G = c * (c + 1) + delta; |
| 483 |
value_type alpha = a + b; |
| 484 |
value_type alpha2 = alpha * alpha; |
| 485 |
value_type eta = (2 * alpha + 1) * (2 * alpha + 1) + 1; |
| 486 |
value_type H = 3 * alpha2 + 5 * alpha + 2; |
| 487 |
value_type F = alpha2 * (alpha + 1) + H * delta |
| 488 |
+ (2 * alpha + 4) * delta2 + delta3; |
| 489 |
value_type P = (3 * alpha + 1) * (9 * alpha + 17) |
| 490 |
+ 2 * alpha * (3 * alpha + 2) * (3 * alpha + 4) + 15; |
| 491 |
value_type Q = 54 * alpha2 + 162 * alpha + 130; |
| 492 |
value_type R = 6 * (6 * alpha + 11); |
| 493 |
value_type D = delta |
| 494 |
* (H * H + 2 * P * delta + Q * delta2 + R * delta3 + 9 * delta4); |
| 495 |
value_type variance = (b / G) |
| 496 |
* (1 + delta * (l * l + 3 * l + eta) / (G * G)) |
| 497 |
- (b * b / F) * (1 + D / (F * F)); |
| 498 |
value_type sd = sqrt(variance); |
| 499 |
|
| 500 |
value_type guess = comp |
| 501 |
? quantile(complement(normal_distribution<RealType, Policy>(static_cast<RealType>(mean), static_cast<RealType>(sd)), p)) |
| 502 |
: quantile(normal_distribution<RealType, Policy>(static_cast<RealType>(mean), static_cast<RealType>(sd)), p); |
| 503 |
|
| 504 |
if(guess >= 1) |
| 505 |
guess = mean; |
| 506 |
if(guess <= tools::min_value<value_type>()) |
| 507 |
guess = mean; |
| 508 |
*/ |
| 509 |
value_type guess = mean; |
| 510 |
detail::nc_beta_quantile_functor<value_type, Policy> |
| 511 |
f(non_central_beta_distribution<value_type, Policy>(a, b, l), p, comp); |
| 512 |
tools::eps_tolerance<value_type> tol(policies::digits<RealType, Policy>()); |
| 513 |
boost::math::uintmax_t max_iter = policies::get_max_root_iterations<Policy>(); |
| 514 |
|
| 515 |
boost::math::pair<value_type, value_type> ir |
| 516 |
= bracket_and_solve_root_01( |
| 517 |
f, guess, value_type(2.5), true, tol, |
| 518 |
max_iter, Policy()); |
| 519 |
value_type result = ir.first + (ir.second - ir.first) / 2; |
| 520 |
|
| 521 |
if(max_iter >= policies::get_max_root_iterations<Policy>()) |
| 522 |
{ |
| 523 |
// LCOV_EXCL_START |
| 524 |
return policies::raise_evaluation_error<RealType>(function, "Unable to locate solution in a reasonable time:" |
| 525 |
" either there is no answer to quantile of the non central beta distribution" |
| 526 |
" or the answer is infinite. Current best guess is %1%", |
| 527 |
policies::checked_narrowing_cast<RealType, forwarding_policy>( |
| 528 |
result, |
| 529 |
function), Policy()); |
| 530 |
// LCOV_EXCL_STOP |
| 531 |
} |
| 532 |
return policies::checked_narrowing_cast<RealType, forwarding_policy>( |
| 533 |
result, |
| 534 |
function); |
| 535 |
} |
| 536 |
|
| 537 |
template <class T, class Policy> |
| 538 |
BOOST_MATH_GPU_ENABLED T non_central_beta_pdf(T a, T b, T lam, T x, T y, const Policy& pol) |
| 539 |
{ |
| 540 |
BOOST_MATH_STD_USING |
| 541 |
// |
| 542 |
// Special cases: |
| 543 |
// |
| 544 |
if((x == 0) || (y == 0)) |
| 545 |
return 0; |
| 546 |
// |
| 547 |
// Variables come first: |
| 548 |
// |
| 549 |
boost::math::uintmax_t max_iter = policies::get_max_series_iterations<Policy>(); |
| 550 |
T errtol = boost::math::policies::get_epsilon<T, Policy>(); |
| 551 |
T l2 = lam / 2; |
| 552 |
// |
| 553 |
// k is the starting point for iteration, and is the |
| 554 |
// maximum of the poisson weighting term: |
| 555 |
// |
| 556 |
long long k = lltrunc(l2); |
| 557 |
// Starting Poisson weight: |
| 558 |
T pois = gamma_p_derivative(T(k+1), l2, pol); |
| 559 |
// Starting beta term: |
| 560 |
T beta = x < y ? |
| 561 |
ibeta_derivative(a + k, b, x, pol) |
| 562 |
: ibeta_derivative(b, a + k, y, pol); |
| 563 |
|
| 564 |
while (fabs(beta * pois) < tools::min_value<T>()) |
| 565 |
{ |
| 566 |
if ((k == 0) || (pois == 0)) |
| 567 |
return 0; // Nothing else we can do! |
| 568 |
// |
| 569 |
// We only get here when a+k and b are large and x is small, |
| 570 |
// in that case reduce k (bisect) until both terms are finite: |
| 571 |
// |
| 572 |
k /= 2; |
| 573 |
pois = gamma_p_derivative(T(k + 1), l2, pol); |
| 574 |
// Starting beta term: |
| 575 |
beta = x < y ? |
| 576 |
ibeta_derivative(a + k, b, x, pol) |
| 577 |
: ibeta_derivative(b, a + k, y, pol); |
| 578 |
} |
| 579 |
|
| 580 |
|
| 581 |
T sum = 0; |
| 582 |
T poisf(pois); |
| 583 |
T betaf(beta); |
| 584 |
|
| 585 |
// |
| 586 |
// Stable backwards recursion first: |
| 587 |
// |
| 588 |
boost::math::uintmax_t count = k; |
| 589 |
T ratio = 0; |
| 590 |
T old_ratio = 0; |
| 591 |
for(auto i = k; i >= 0; --i) |
| 592 |
{ |
| 593 |
T term = beta * pois; |
| 594 |
sum += term; |
| 595 |
ratio = fabs(term / sum); |
| 596 |
if(((ratio < errtol) && (ratio < old_ratio)) || (term == 0)) |
| 597 |
{ |
| 598 |
count = k - i; |
| 599 |
break; |
| 600 |
} |
| 601 |
ratio = old_ratio; |
| 602 |
pois *= i / l2; |
| 603 |
|
| 604 |
if (a + b + i != 1) |
| 605 |
{ |
| 606 |
beta *= (a + i - 1) / (x * (a + i + b - 1)); |
| 607 |
} |
| 608 |
} |
| 609 |
old_ratio = 0; |
| 610 |
for(auto i = k + 1; ; ++i) |
| 611 |
{ |
| 612 |
poisf *= l2 / i; |
| 613 |
betaf *= x * (a + b + i - 1) / (a + i - 1); |
| 614 |
|
| 615 |
T term = poisf * betaf; |
| 616 |
sum += term; |
| 617 |
ratio = fabs(term / sum); |
| 618 |
if(((ratio < errtol) && (ratio < old_ratio)) || (term == 0)) |
| 619 |
{ |
| 620 |
break; |
| 621 |
} |
| 622 |
old_ratio = ratio; |
| 623 |
if(static_cast<boost::math::uintmax_t>(count + i - k) > max_iter) |
| 624 |
{ |
| 625 |
return policies::raise_evaluation_error("pdf(non_central_beta_distribution<%1%>, %1%)", "Series did not converge, closest value was %1%", sum, pol); // LCOV_EXCL_LINE |
| 626 |
} |
| 627 |
} |
| 628 |
return sum; |
| 629 |
} |
| 630 |
|
| 631 |
template <class RealType, class Policy> |
| 632 |
BOOST_MATH_GPU_ENABLED RealType nc_beta_pdf(const non_central_beta_distribution<RealType, Policy>& dist, const RealType& x) |
| 633 |
{ |
| 634 |
BOOST_MATH_STD_USING |
| 635 |
constexpr auto function = "pdf(non_central_beta_distribution<%1%>, %1%)"; |
| 636 |
typedef typename policies::evaluation<RealType, Policy>::type value_type; |
| 637 |
typedef typename policies::normalise< |
| 638 |
Policy, |
| 639 |
policies::promote_float<false>, |
| 640 |
policies::promote_double<false>, |
| 641 |
policies::discrete_quantile<>, |
| 642 |
policies::assert_undefined<> >::type forwarding_policy; |
| 643 |
|
| 644 |
value_type a = dist.alpha(); |
| 645 |
value_type b = dist.beta(); |
| 646 |
value_type l = dist.non_centrality(); |
| 647 |
value_type r; |
| 648 |
if(!beta_detail::check_alpha( |
| 649 |
function, |
| 650 |
a, &r, Policy()) |
| 651 |
|| |
| 652 |
!beta_detail::check_beta( |
| 653 |
function, |
| 654 |
b, &r, Policy()) |
| 655 |
|| |
| 656 |
!detail::check_non_centrality( |
| 657 |
function, |
| 658 |
l, |
| 659 |
&r, |
| 660 |
Policy()) |
| 661 |
|| |
| 662 |
!beta_detail::check_x( |
| 663 |
function, |
| 664 |
static_cast<value_type>(x), |
| 665 |
&r, |
| 666 |
Policy())) |
| 667 |
return static_cast<RealType>(r); |
| 668 |
|
| 669 |
if(l == 0) |
| 670 |
return pdf(boost::math::beta_distribution<RealType, Policy>(dist.alpha(), dist.beta()), x); |
| 671 |
return policies::checked_narrowing_cast<RealType, forwarding_policy>( |
| 672 |
non_central_beta_pdf(a, b, l, static_cast<value_type>(x), value_type(1 - static_cast<value_type>(x)), forwarding_policy()), |
| 673 |
"function"); |
| 674 |
} |
| 675 |
|
| 676 |
template <class T> |
| 677 |
struct hypergeometric_2F2_sum |
| 678 |
{ |
| 679 |
typedef T result_type; |
| 680 |
BOOST_MATH_GPU_ENABLED hypergeometric_2F2_sum(T a1_, T a2_, T b1_, T b2_, T z_) : a1(a1_), a2(a2_), b1(b1_), b2(b2_), z(z_), term(1), k(0) {} |
| 681 |
BOOST_MATH_GPU_ENABLED T operator()() |
| 682 |
{ |
| 683 |
T result = term; |
| 684 |
term *= a1 * a2 / (b1 * b2); |
| 685 |
a1 += 1; |
| 686 |
a2 += 1; |
| 687 |
b1 += 1; |
| 688 |
b2 += 1; |
| 689 |
k += 1; |
| 690 |
term /= k; |
| 691 |
term *= z; |
| 692 |
return result; |
| 693 |
} |
| 694 |
T a1, a2, b1, b2, z, term, k; |
| 695 |
}; |
| 696 |
|
| 697 |
template <class T, class Policy> |
| 698 |
BOOST_MATH_GPU_ENABLED T hypergeometric_2F2(T a1, T a2, T b1, T b2, T z, const Policy& pol) |
| 699 |
{ |
| 700 |
typedef typename policies::evaluation<T, Policy>::type value_type; |
| 701 |
|
| 702 |
const char* function = "boost::math::detail::hypergeometric_2F2<%1%>(%1%,%1%,%1%,%1%,%1%)"; |
| 703 |
|
| 704 |
hypergeometric_2F2_sum<value_type> s(a1, a2, b1, b2, z); |
| 705 |
boost::math::uintmax_t max_iter = policies::get_max_series_iterations<Policy>(); |
| 706 |
|
| 707 |
value_type result = boost::math::tools::sum_series(s, boost::math::policies::get_epsilon<value_type, Policy>(), max_iter); |
| 708 |
|
| 709 |
policies::check_series_iterations<T>(function, max_iter, pol); |
| 710 |
return policies::checked_narrowing_cast<T, Policy>(result, function); |
| 711 |
} |
| 712 |
|
| 713 |
} // namespace detail |
| 714 |
|
| 715 |
template <class RealType = double, class Policy = policies::policy<> > |
| 716 |
class non_central_beta_distribution |
| 717 |
{ |
| 718 |
public: |
| 719 |
typedef RealType value_type; |
| 720 |
typedef Policy policy_type; |
| 721 |
|
| 722 |
BOOST_MATH_GPU_ENABLED non_central_beta_distribution(RealType a_, RealType b_, RealType lambda) : a(a_), b(b_), ncp(lambda) |
| 723 |
{ |
| 724 |
const char* function = "boost::math::non_central_beta_distribution<%1%>::non_central_beta_distribution(%1%,%1%)"; |
| 725 |
RealType r; |
| 726 |
beta_detail::check_alpha( |
| 727 |
function, |
| 728 |
a, &r, Policy()); |
| 729 |
beta_detail::check_beta( |
| 730 |
function, |
| 731 |
b, &r, Policy()); |
| 732 |
detail::check_non_centrality( |
| 733 |
function, |
| 734 |
lambda, |
| 735 |
&r, |
| 736 |
Policy()); |
| 737 |
} // non_central_beta_distribution constructor. |
| 738 |
|
| 739 |
BOOST_MATH_GPU_ENABLED RealType alpha() const |
| 740 |
{ // Private data getter function. |
| 741 |
return a; |
| 742 |
} |
| 743 |
BOOST_MATH_GPU_ENABLED RealType beta() const |
| 744 |
{ // Private data getter function. |
| 745 |
return b; |
| 746 |
} |
| 747 |
BOOST_MATH_GPU_ENABLED RealType non_centrality() const |
| 748 |
{ // Private data getter function. |
| 749 |
return ncp; |
| 750 |
} |
| 751 |
private: |
| 752 |
// Data member, initialized by constructor. |
| 753 |
RealType a; // alpha. |
| 754 |
RealType b; // beta. |
| 755 |
RealType ncp; // non-centrality parameter |
| 756 |
}; // template <class RealType, class Policy> class non_central_beta_distribution |
| 757 |
|
| 758 |
typedef non_central_beta_distribution<double> non_central_beta; // Reserved name of type double. |
| 759 |
|
| 760 |
#ifdef __cpp_deduction_guides |
| 761 |
template <class RealType> |
| 762 |
non_central_beta_distribution(RealType,RealType,RealType)->non_central_beta_distribution<typename boost::math::tools::promote_args<RealType>::type>; |
| 763 |
#endif |
| 764 |
|
| 765 |
// Non-member functions to give properties of the distribution. |
| 766 |
|
| 767 |
template <class RealType, class Policy> |
| 768 |
BOOST_MATH_GPU_ENABLED inline const boost::math::pair<RealType, RealType> range(const non_central_beta_distribution<RealType, Policy>& /* dist */) |
| 769 |
{ // Range of permissible values for random variable k. |
| 770 |
using boost::math::tools::max_value; |
| 771 |
return boost::math::pair<RealType, RealType>(static_cast<RealType>(0), static_cast<RealType>(1)); |
| 772 |
} |
| 773 |
|
| 774 |
template <class RealType, class Policy> |
| 775 |
BOOST_MATH_GPU_ENABLED inline const boost::math::pair<RealType, RealType> support(const non_central_beta_distribution<RealType, Policy>& /* dist */) |
| 776 |
{ // Range of supported values for random variable k. |
| 777 |
// This is range where cdf rises from 0 to 1, and outside it, the pdf is zero. |
| 778 |
using boost::math::tools::max_value; |
| 779 |
return boost::math::pair<RealType, RealType>(static_cast<RealType>(0), static_cast<RealType>(1)); |
| 780 |
} |
| 781 |
|
| 782 |
template <class RealType, class Policy> |
| 783 |
BOOST_MATH_GPU_ENABLED inline RealType mode(const non_central_beta_distribution<RealType, Policy>& dist) |
| 784 |
{ // mode. |
| 785 |
constexpr auto function = "mode(non_central_beta_distribution<%1%> const&)"; |
| 786 |
|
| 787 |
RealType a = dist.alpha(); |
| 788 |
RealType b = dist.beta(); |
| 789 |
RealType l = dist.non_centrality(); |
| 790 |
RealType r; |
| 791 |
if(!beta_detail::check_alpha( |
| 792 |
function, |
| 793 |
a, &r, Policy()) |
| 794 |
|| |
| 795 |
!beta_detail::check_beta( |
| 796 |
function, |
| 797 |
b, &r, Policy()) |
| 798 |
|| |
| 799 |
!detail::check_non_centrality( |
| 800 |
function, |
| 801 |
l, |
| 802 |
&r, |
| 803 |
Policy())) |
| 804 |
return static_cast<RealType>(r); |
| 805 |
RealType c = a + b + l / 2; |
| 806 |
RealType mean = 1 - (b / c) * (1 + l / (2 * c * c)); |
| 807 |
return detail::generic_find_mode_01( |
| 808 |
dist, |
| 809 |
mean, |
| 810 |
function); |
| 811 |
} |
| 812 |
|
| 813 |
// |
| 814 |
// We don't have the necessary information to implement |
| 815 |
// these at present. These are just disabled for now, |
| 816 |
// prototypes retained so we can fill in the blanks |
| 817 |
// later: |
| 818 |
// |
| 819 |
template <class RealType, class Policy> |
| 820 |
BOOST_MATH_GPU_ENABLED inline RealType mean(const non_central_beta_distribution<RealType, Policy>& dist) |
| 821 |
{ |
| 822 |
BOOST_MATH_STD_USING |
| 823 |
RealType a = dist.alpha(); |
| 824 |
RealType b = dist.beta(); |
| 825 |
RealType d = dist.non_centrality(); |
| 826 |
RealType apb = a + b; |
| 827 |
return exp(-d / 2) * a * detail::hypergeometric_2F2<RealType, Policy>(1 + a, apb, a, 1 + apb, d / 2, Policy()) / apb; |
| 828 |
} // mean |
| 829 |
|
| 830 |
template <class RealType, class Policy> |
| 831 |
BOOST_MATH_GPU_ENABLED inline RealType variance(const non_central_beta_distribution<RealType, Policy>& dist) |
| 832 |
{ |
| 833 |
// |
| 834 |
// Relative error of this function may be arbitrarily large... absolute |
| 835 |
// error will be small however... that's the best we can do for now. |
| 836 |
// |
| 837 |
BOOST_MATH_STD_USING |
| 838 |
RealType a = dist.alpha(); |
| 839 |
RealType b = dist.beta(); |
| 840 |
RealType d = dist.non_centrality(); |
| 841 |
RealType apb = a + b; |
| 842 |
RealType result = detail::hypergeometric_2F2(RealType(1 + a), apb, a, RealType(1 + apb), RealType(d / 2), Policy()); |
| 843 |
result *= result * -exp(-d) * a * a / (apb * apb); |
| 844 |
result += exp(-d / 2) * a * (1 + a) * detail::hypergeometric_2F2(RealType(2 + a), apb, a, RealType(2 + apb), RealType(d / 2), Policy()) / (apb * (1 + apb)); |
| 845 |
return result; |
| 846 |
} |
| 847 |
|
| 848 |
// RealType standard_deviation(const non_central_beta_distribution<RealType, Policy>& dist) |
| 849 |
// standard_deviation provided by derived accessors. |
| 850 |
template <class RealType, class Policy> |
| 851 |
BOOST_MATH_GPU_ENABLED inline RealType skewness(const non_central_beta_distribution<RealType, Policy>& /*dist*/) |
| 852 |
{ // skewness = sqrt(l). |
| 853 |
const char* function = "boost::math::non_central_beta_distribution<%1%>::skewness()"; |
| 854 |
typedef typename Policy::assert_undefined_type assert_type; |
| 855 |
static_assert(assert_type::value == 0, "The Non Central Beta Distribution has no skewness."); |
| 856 |
|
| 857 |
return policies::raise_evaluation_error<RealType>(function, "This function is not yet implemented, the only sensible result is %1%.", // LCOV_EXCL_LINE |
| 858 |
boost::math::numeric_limits<RealType>::quiet_NaN(), Policy()); // infinity? LCOV_EXCL_LINE |
| 859 |
} |
| 860 |
|
| 861 |
template <class RealType, class Policy> |
| 862 |
BOOST_MATH_GPU_ENABLED inline RealType kurtosis_excess(const non_central_beta_distribution<RealType, Policy>& /*dist*/) |
| 863 |
{ |
| 864 |
const char* function = "boost::math::non_central_beta_distribution<%1%>::kurtosis_excess()"; |
| 865 |
typedef typename Policy::assert_undefined_type assert_type; |
| 866 |
static_assert(assert_type::value == 0, "The Non Central Beta Distribution has no kurtosis excess."); |
| 867 |
|
| 868 |
return policies::raise_evaluation_error<RealType>(function, "This function is not yet implemented, the only sensible result is %1%.", // LCOV_EXCL_LINE |
| 869 |
boost::math::numeric_limits<RealType>::quiet_NaN(), Policy()); // infinity? LCOV_EXCL_LINE |
| 870 |
} // kurtosis_excess |
| 871 |
|
| 872 |
template <class RealType, class Policy> |
| 873 |
BOOST_MATH_GPU_ENABLED inline RealType kurtosis(const non_central_beta_distribution<RealType, Policy>& dist) |
| 874 |
{ |
| 875 |
return kurtosis_excess(dist) + 3; |
| 876 |
} |
| 877 |
|
| 878 |
template <class RealType, class Policy> |
| 879 |
BOOST_MATH_GPU_ENABLED inline RealType pdf(const non_central_beta_distribution<RealType, Policy>& dist, const RealType& x) |
| 880 |
{ // Probability Density/Mass Function. |
| 881 |
return detail::nc_beta_pdf(dist, x); |
| 882 |
} // pdf |
| 883 |
|
| 884 |
template <class RealType, class Policy> |
| 885 |
BOOST_MATH_GPU_ENABLED RealType cdf(const non_central_beta_distribution<RealType, Policy>& dist, const RealType& x) |
| 886 |
{ |
| 887 |
const char* function = "boost::math::non_central_beta_distribution<%1%>::cdf(%1%)"; |
| 888 |
RealType a = dist.alpha(); |
| 889 |
RealType b = dist.beta(); |
| 890 |
RealType l = dist.non_centrality(); |
| 891 |
RealType r; |
| 892 |
if(!beta_detail::check_alpha( |
| 893 |
function, |
| 894 |
a, &r, Policy()) |
| 895 |
|| |
| 896 |
!beta_detail::check_beta( |
| 897 |
function, |
| 898 |
b, &r, Policy()) |
| 899 |
|| |
| 900 |
!detail::check_non_centrality( |
| 901 |
function, |
| 902 |
l, |
| 903 |
&r, |
| 904 |
Policy()) |
| 905 |
|| |
| 906 |
!beta_detail::check_x( |
| 907 |
function, |
| 908 |
x, |
| 909 |
&r, |
| 910 |
Policy())) |
| 911 |
return static_cast<RealType>(r); |
| 912 |
|
| 913 |
if(l == 0) |
| 914 |
return cdf(beta_distribution<RealType, Policy>(a, b), x); |
| 915 |
|
| 916 |
return detail::non_central_beta_cdf(x, RealType(1 - x), a, b, l, false, Policy()); |
| 917 |
} // cdf |
| 918 |
|
| 919 |
template <class RealType, class Policy> |
| 920 |
BOOST_MATH_GPU_ENABLED RealType cdf(const complemented2_type<non_central_beta_distribution<RealType, Policy>, RealType>& c) |
| 921 |
{ // Complemented Cumulative Distribution Function |
| 922 |
const char* function = "boost::math::non_central_beta_distribution<%1%>::cdf(%1%)"; |
| 923 |
non_central_beta_distribution<RealType, Policy> const& dist = c.dist; |
| 924 |
RealType a = dist.alpha(); |
| 925 |
RealType b = dist.beta(); |
| 926 |
RealType l = dist.non_centrality(); |
| 927 |
RealType x = c.param; |
| 928 |
RealType r; |
| 929 |
if(!beta_detail::check_alpha( |
| 930 |
function, |
| 931 |
a, &r, Policy()) |
| 932 |
|| |
| 933 |
!beta_detail::check_beta( |
| 934 |
function, |
| 935 |
b, &r, Policy()) |
| 936 |
|| |
| 937 |
!detail::check_non_centrality( |
| 938 |
function, |
| 939 |
l, |
| 940 |
&r, |
| 941 |
Policy()) |
| 942 |
|| |
| 943 |
!beta_detail::check_x( |
| 944 |
function, |
| 945 |
x, |
| 946 |
&r, |
| 947 |
Policy())) |
| 948 |
return static_cast<RealType>(r); |
| 949 |
|
| 950 |
if(l == 0) |
| 951 |
return cdf(complement(beta_distribution<RealType, Policy>(a, b), x)); |
| 952 |
|
| 953 |
return detail::non_central_beta_cdf(x, RealType(1 - x), a, b, l, true, Policy()); |
| 954 |
} // ccdf |
| 955 |
|
| 956 |
template <class RealType, class Policy> |
| 957 |
BOOST_MATH_GPU_ENABLED inline RealType quantile(const non_central_beta_distribution<RealType, Policy>& dist, const RealType& p) |
| 958 |
{ // Quantile (or Percent Point) function. |
| 959 |
return detail::nc_beta_quantile(dist, p, false); |
| 960 |
} // quantile |
| 961 |
|
| 962 |
template <class RealType, class Policy> |
| 963 |
BOOST_MATH_GPU_ENABLED inline RealType quantile(const complemented2_type<non_central_beta_distribution<RealType, Policy>, RealType>& c) |
| 964 |
{ // Quantile (or Percent Point) function. |
| 965 |
return detail::nc_beta_quantile(c.dist, c.param, true); |
| 966 |
} // quantile complement. |
| 967 |
|
| 968 |
} // namespace math |
| 969 |
} // namespace boost |
| 970 |
|
| 971 |
// This include must be at the end, *after* the accessors |
| 972 |
// for this distribution have been defined, in order to |
| 973 |
// keep compilers that support two-phase lookup happy. |
| 974 |
#include <boost/math/distributions/detail/derived_accessors.hpp> |
| 975 |
|
| 976 |
#endif // BOOST_MATH_SPECIAL_NON_CENTRAL_BETA_HPP |
| 977 |
|