1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 2008 Free Software Foundation, Inc.
4 This program is free software: you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation, either version 3 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program. If not, see <http://www.gnu.org/licenses/>. */
19 #include <gsl/gsl_roots.h>
20 #include <gsl/gsl_sf.h>
21 #include <libpspp/assertion.h>
22 #include <libpspp/pool.h>
25 const struct substring empty_string = {NULL, 0};
28 expr_error (void *aux UNUSED, const char *format, ...)
33 m.category = MSG_C_SYNTAX;
34 m.severity = MSG_S_ERROR;
35 va_start (args, format);
36 m.text = xvasprintf (format, args);
43 expr_ymd_to_ofs (double year, double month, double day)
49 if (y != year || m != month || d != day)
51 msg (SE, _("One of the arguments to a DATE function is not an integer. "
52 "The result will be system-missing."));
56 return calendar_gregorian_to_offset (y, m, d, expr_error, NULL);
60 expr_ymd_to_date (double year, double month, double day)
62 double ofs = expr_ymd_to_ofs (year, month, day);
63 return ofs != SYSMIS ? ofs * DAY_S : SYSMIS;
67 expr_wkyr_to_date (double week, double year)
73 msg (SE, _("The week argument to DATE.WKYR is not an integer. "
74 "The result will be system-missing."));
77 else if (w < 1 || w > 53)
79 msg (SE, _("The week argument to DATE.WKYR is outside the acceptable "
81 "The result will be system-missing."));
86 double yr_1_1 = expr_ymd_to_ofs (year, 1, 1);
88 return DAY_S * (yr_1_1 + WEEK_DAY * (w - 1));
95 expr_yrday_to_date (double year, double yday)
101 msg (SE, _("The day argument to DATE.YRDAY is not an integer. "
102 "The result will be system-missing."));
105 else if (yd < 1 || yd > 366)
107 msg (SE, _("The day argument to DATE.YRDAY is outside the acceptable "
108 "range of 1 to 366. "
109 "The result will be system-missing."));
114 double yr_1_1 = expr_ymd_to_ofs (year, 1, 1);
115 if (yr_1_1 != SYSMIS)
116 return DAY_S * (yr_1_1 + yd - 1.);
123 expr_yrmoda (double year, double month, double day)
125 if (year >= 0 && year <= 99)
127 else if (year != (int) year && year > 47516)
129 msg (SE, _("The year argument to YRMODA is greater than 47516. "
130 "The result will be system-missing."));
134 return expr_ymd_to_ofs (year, month, day);
150 /* Stores in *UNIT the unit whose name is NAME.
152 static enum date_unit
153 recognize_unit (struct substring name, enum date_unit *unit)
158 const struct substring name;
160 static const struct unit_name unit_names[] =
162 { DATE_YEARS, SS_LITERAL_INITIALIZER ("years") },
163 { DATE_QUARTERS, SS_LITERAL_INITIALIZER ("quarters") },
164 { DATE_MONTHS, SS_LITERAL_INITIALIZER ("months") },
165 { DATE_WEEKS, SS_LITERAL_INITIALIZER ("weeks") },
166 { DATE_DAYS, SS_LITERAL_INITIALIZER ("days") },
167 { DATE_HOURS, SS_LITERAL_INITIALIZER ("hours") },
168 { DATE_MINUTES, SS_LITERAL_INITIALIZER ("minutes") },
169 { DATE_SECONDS, SS_LITERAL_INITIALIZER ("seconds") },
171 const int unit_name_cnt = sizeof unit_names / sizeof *unit_names;
173 const struct unit_name *un;
175 for (un = unit_names; un < &unit_names[unit_name_cnt]; un++)
176 if (ss_equals_case (un->name, name))
182 /* TRANSLATORS: Don't translate the the actual unit names `weeks', `days' etc
183 They must remain in their original English. */
184 msg (SE, _("Unrecognized date unit `%.*s'. "
185 "Valid date units are `years', `quarters', `months', "
186 "`weeks', `days', `hours', `minutes', and `seconds'."),
187 (int) ss_length (name), ss_data (name));
191 /* Returns the number of whole years from DATE1 to DATE2,
192 where a year is defined as the same or later month, day, and
195 year_diff (double date1, double date2)
201 assert (date2 >= date1);
202 calendar_offset_to_gregorian (date1 / DAY_S, &y1, &m1, &d1, &yd1);
203 calendar_offset_to_gregorian (date2 / DAY_S, &y2, &m2, &d2, &yd2);
208 int yd1 = 32 * m1 + d1;
209 int yd2 = 32 * m2 + d2;
211 || (yd2 == yd1 && fmod (date2, DAY_S) < fmod (date1, DAY_S)))
217 /* Returns the number of whole months from DATE1 to DATE2,
218 where a month is defined as the same or later day and time of
221 month_diff (double date1, double date2)
227 assert (date2 >= date1);
228 calendar_offset_to_gregorian (date1 / DAY_S, &y1, &m1, &d1, &yd1);
229 calendar_offset_to_gregorian (date2 / DAY_S, &y2, &m2, &d2, &yd2);
231 diff = ((y2 * 12) + m2) - ((y1 * 12) + m1);
234 || (d2 == d1 && fmod (date2, DAY_S) < fmod (date1, DAY_S))))
239 /* Returns the number of whole quarter from DATE1 to DATE2,
240 where a quarter is defined as three months. */
242 quarter_diff (double date1, double date2)
244 return month_diff (date1, date2) / 3;
247 /* Returns the number of seconds in the given UNIT. */
249 date_unit_duration (enum date_unit unit)
273 /* Returns the span from DATE1 to DATE2 in terms of UNIT_NAME. */
275 expr_date_difference (double date1, double date2, struct substring unit_name)
279 if (!recognize_unit (unit_name, &unit))
285 return (date2 >= date1
286 ? year_diff (date1, date2)
287 : -year_diff (date2, date1));
290 return (date2 >= date1
291 ? quarter_diff (date1, date2)
292 : -quarter_diff (date2, date1));
295 return (date2 >= date1
296 ? month_diff (date1, date2)
297 : -month_diff (date2, date1));
304 return trunc ((date2 - date1) / date_unit_duration (unit));
310 /* How to deal with days out of range for a given month. */
313 SUM_ROLLOVER, /* Roll them over to the next month. */
314 SUM_CLOSEST /* Use the last day of the month. */
317 /* Stores in *METHOD the method whose name is NAME.
320 recognize_method (struct substring method_name, enum date_sum_method *method)
322 if (ss_equals_case (method_name, ss_cstr ("closest")))
324 *method = SUM_CLOSEST;
327 else if (ss_equals_case (method_name, ss_cstr ("rollover")))
329 *method = SUM_ROLLOVER;
334 msg (SE, _("Invalid DATESUM method. "
335 "Valid choices are `closest' and `rollover'."));
340 /* Returns DATE advanced by the given number of MONTHS, with
341 day-of-month overflow resolved using METHOD. */
343 add_months (double date, int months, enum date_sum_method method)
348 calendar_offset_to_gregorian (date / DAY_S, &y, &m, &d, &yd);
361 assert (m >= 1 && m <= 12);
363 if (method == SUM_CLOSEST && d > calendar_days_in_month (y, m))
364 d = calendar_days_in_month (y, m);
366 output = calendar_gregorian_to_offset (y, m, d, expr_error, NULL);
367 if (output != SYSMIS)
368 output = (output * DAY_S) + fmod (date, DAY_S);
372 /* Returns DATE advanced by the given QUANTITY of units given in
373 UNIT_NAME, with day-of-month overflow resolved using
376 expr_date_sum (double date, double quantity, struct substring unit_name,
377 struct substring method_name)
380 enum date_sum_method method;
382 if (!recognize_unit (unit_name, &unit)
383 || !recognize_method (method_name, &method))
389 return add_months (date, trunc (quantity) * 12, method);
392 return add_months (date, trunc (quantity) * 3, method);
395 return add_months (date, trunc (quantity), method);
402 return date + quantity * date_unit_duration (unit);
409 compare_string_3way (const struct substring *a, const struct substring *b)
413 for (i = 0; i < a->length && i < b->length; i++)
414 if (a->string[i] != b->string[i])
415 return a->string[i] < b->string[i] ? -1 : 1;
416 for (; i < a->length; i++)
417 if (a->string[i] != ' ')
419 for (; i < b->length; i++)
420 if (b->string[i] != ' ')
426 count_valid (double *d, size_t d_cnt)
432 for (i = 0; i < d_cnt; i++)
433 valid_cnt += is_valid (d[i]);
438 alloc_string (struct expression *e, size_t length)
442 s.string = pool_alloc (e->eval_pool, length);
447 copy_string (struct expression *e, const char *old, size_t length)
449 struct substring s = alloc_string (e, length);
450 memcpy (s.string, old, length);
454 /* Returns the noncentral beta cumulative distribution function
455 value for the given arguments.
457 FIXME: The accuracy of this function is not entirely
458 satisfactory. We only match the example values given in AS
459 310 to the first 5 significant digits. */
461 ncdf_beta (double x, double a, double b, double lambda)
465 if (x <= 0. || x >= 1. || a <= 0. || b <= 0. || lambda <= 0.)
471 /* Algorithm AS 226. */
472 double x0, a0, beta, temp, gx, q, ax, sumq, sum;
473 double err_max = 2 * DBL_EPSILON;
478 x0 = floor (c - 5.0 * sqrt (c));
482 beta = (gsl_sf_lngamma (a0)
484 - gsl_sf_lngamma (a0 + b));
485 temp = gsl_sf_beta_inc (a0, b, x);
486 gx = exp (a0 * log (x) + b * log (1. - x) - beta - log (a0));
488 q = exp (-c + x0 * log (c)) - gsl_sf_lngamma (x0 + 1.);
500 gx = x * (a + b + iter - 1.) * gx / (a + iter);
506 err_bound = (temp - gx) * sumq;
508 while (iter < iter_max && err_bound > err_max);
514 /* Algorithm AS 310. */
516 int iter, iter_lower, iter_upper, iter1, iter2, j;
517 double t, q, r, psum, beta, s1, gx, fx, temp, ftemp, t0, s0, sum, s;
519 double err_max = 2 * DBL_EPSILON;
525 iter_lower = m - 5. * m_sqrt;
526 iter_upper = m + 5. * m_sqrt;
528 t = -c + m * log (c) - gsl_sf_lngamma (m + 1.);
532 beta = (gsl_sf_lngamma (a + m)
534 - gsl_sf_lngamma (a + m + b));
535 s1 = (a + m) * log (x) + b * log (1. - x) - log (a + m) - beta;
537 ftemp = temp = gsl_sf_beta_inc (a + m, b, x);
542 while (iter1 >= iter_lower && q >= err_max)
546 gx = (a + iter1) / (x * (a + b + iter1 - 1.)) * gx;
553 t0 = (gsl_sf_lngamma (a + b)
554 - gsl_sf_lngamma (a + 1.)
555 - gsl_sf_lngamma (b));
556 s0 = a * log (x) + b * log (1. - x);
559 for (j = 0; j < iter1; j++)
562 s += exp (t0 + s0 + j * log (x));
563 t1 = log (a + b + j) - log (a + 1. + j) + t0;
567 err_bound = (1. - gsl_sf_gamma_inc_P (iter1, c)) * (temp + s);
574 double ebd = err_bound + (1. - psum) * temp;
575 if (ebd < err_max || iter >= iter_upper)
583 gx = x * (a + b + iter2 - 1.) / (a + iter2) * gx;
592 cdf_bvnor (double x0, double x1, double r)
594 double z = pow2 (x0) - 2. * r * x0 * x1 + pow2 (x1);
595 return exp (-z / (2. * (1 - r * r))) * (2. * M_PI * sqrt (1 - r * r));
599 idf_fdist (double P, double df1, double df2)
601 double temp = gsl_cdf_beta_Pinv (P, df1 / 2, df2 / 2);
602 return temp * df2 / ((1. - temp) * df1);
606 * Mathlib : A C Library of Special Functions
607 * Copyright (C) 1998 Ross Ihaka
608 * Copyright (C) 2000 The R Development Core Team
610 * This program is free software; you can redistribute it and/or
612 * it under the terms of the GNU General Public License as
614 * the Free Software Foundation; either version 2 of the
616 * (at your option) any later version.
618 * This program is distributed in the hope that it will be
620 * but WITHOUT ANY WARRANTY; without even the implied warranty
622 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
623 * GNU General Public License for more details.
625 * You should have received a copy of the GNU General Public
627 * along with this program; if not, write to the Free Software
628 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
632 /* Returns the density of the noncentral beta distribution with
633 noncentrality parameter LAMBDA. */
635 npdf_beta (double x, double a, double b, double lambda)
637 if (lambda < 0. || a <= 0. || b <= 0.)
639 else if (lambda == 0.)
640 return gsl_ran_beta_pdf (x, a, b);
643 double max_error = 2 * DBL_EPSILON;
645 double term = gsl_ran_beta_pdf (x, a, b);
646 double lambda2 = 0.5 * lambda;
647 double weight = exp (-lambda2);
648 double sum = weight * term;
649 double psum = weight;
651 for (k = 1; k <= max_iter && 1 - psum < max_error; k++) {
652 weight *= lambda2 / k;
653 term *= x * (a + b) / a;
654 sum += weight * term;