Talk:Closest-pair problem/C

your code does NOT RUN when i compile with devc or Turbo C???? (unsigned comment added by 113.22.126.190 at 21:06, 24 October 2010)

Correct it? Try setting compiler flags for compatibility with a specific C standard? --Michael Mol 13:11, 25 October 2010 (UTC)

"The code does not run"... is like pretending to derive solar physics by the statement "the sun shines"... By the way "devc" (Dev C++ IDE?) usually is used with (an old version of) gcc. On my test machine (GNU/Linux) it runs, except for some "evil dataset" that someone gave me once upon a time... I've inspected the code with valgrind, debugged it, ... but I was not able to unwind the flow that gives the problem... I know this code hides some oddity somewhere. Likely the better thing is to rewrite it from scratch, but I've not the courage yet! :) — ShinTakezou 17:52, 30 May 2011 (UTC)

Propose replacing code

I suggest replace current code sample with rewritten code below. Reasons: 1. It doesn't segfault with 200,000 points or more; 2. It's shorter and quite a bit faster; 3. It's cleaner IMO. <lang C>#include <stdio.h>

1. include <stdlib.h>
2. include <values.h>
3. include <math.h>

typedef struct { double x, y; } point_t, *point;

/* note: even though l_list and r_list are used by each recursion of closest(), they

  are always used in sequence, no need to allocate them repeatedly */

int *l_list, *r_list;

inline double dist(point a, point b) {

       double dx = a->x - b->x, dy = a->y - b->y;
       return dx * dx + dy * dy;

}

inline int cmp_dbl(double a, double b) {

       return a < b ? -1 : a > b ? 1 : 0;

}

double brute_force(point pts, int max_n, int *a, int *b) {

       int i, j;
       point pi;
       double d, min_dist = MAXDOUBLE;

       for (i = 0, pi = pts; i < max_n; pi++, i++) {
               for (j = i + 1; j < max_n; j++) {
                       if ( (d = dist(pi, pts + j)) < min_dist ) {
                               *a = i;
                               *b = j;
                               min_dist = d;
                       }
               }
       }
       return min_dist;

}

void sort_y(point pts, int n, int *list) {

       int cmp_y(const void *a, const void *b) {
               return cmp_dbl(pts[*(int*)a].y, pts[*(int*)b].y);
       }
       qsort(list, n, sizeof(int), cmp_y);

}

double closest(point pts, int n, point *a, point *b) {

       int left, right, lsize, rsize, i;
       point pl, a1, b1;
       double min_d, d, dsqrt, median;

       /* don't divide if there aren't enough points to divide */
       if (n == 2) {
               *a = pts;
               *b = pts + 1;
               return dist(pts, pts + 1);
       }

       if (n == 3) {
               min_d = dist(pts, pts + 1);
               *a = pts; *b = pts + 1;

               if (min_d > (d = dist(pts, pts + 2))) {
                       min_d = d;
                       *b = pts + 2;
               }
               if (min_d > (d = dist(pts + 1, pts + 2))) {
                       min_d = d;
                       *a = pts + 1;
                       *b = pts + 2;
               }
               return min_d;
       }

       /* get left and right results */
       left = right = n / 2;
       if ((min_d = closest(pts, left, a, b)) > (d = closest(pts + left, n - left, &a1, &b1))) {
               min_d = d;
               *a = a1;
               *b = b1;
       }

       median = pts[left].x;
       dsqrt = sqrt(min_d);

       /* find points within +- min distance on X from the center line, list their indices */
       for (   lsize = 0, left--;
               median - pts[left].x  < dsqrt && left;
               l_list[lsize++] = left--);

       for (   rsize = 0;
               pts[right].x - median < dsqrt && right < n;
               r_list[rsize++] = right++);

       /* sort the indices by y: don't touch the point data which is sorted by x */
       sort_y(pts, lsize, l_list);
       sort_y(pts, rsize, r_list);

       /* climb up left and right list and compare distance */
       for (left = right = 0; left < lsize; left ++) {
               /* next point in left list */
               pl = pts + l_list[left];
               median = pl->y;

               /* climb up right list until the y is not too low */
               while (median > dsqrt + pts[r_list[right]].y && right < rsize)
                       right++;

               if (right >= rsize) break;

               for (i = right; i < rsize; i++) {
                       /* right y is too high, break */
                       if (pts[ r_list[i] ].y > median + dsqrt) break;

                       if (min_d > (d = dist(pts + r_list[i], pl))) {
                               *a = pl;
                               *b = pts + r_list[i];
                               min_d = d;
                               dsqrt = sqrt(min_d);
                       }
               }
       }

       return min_d;

}

1. define NP 1000000

int main(int argc, char **argv) {

       int i, j;
       double d;
       point a, b;

       point pts = malloc(sizeof(point_t) * NP);
       l_list = malloc(sizeof(int) * NP);
       r_list = malloc(sizeof(int) * NP);

       for(i = 0; i < NP; i++) {
               pts[i].x = 20 * (double) rand()/RAND_MAX;
               pts[i].y = 20 * (double) rand()/RAND_MAX;
       }

       d = brute_force(pts, NP, &i, &j);
       printf("brute force: %g, between (%f,%f) and (%f,%f)\n",
                       d, pts[i].x, pts[i].y, pts[j].x, pts[j].y);

       int cmp_x(const void *a, const void *b) {
               return cmp_dbl( ((point)a)->x, ((point)b)->x );
       }
       qsort(pts, NP, sizeof(point_t), cmp_x);

       printf("min: %g; ", sqrt(closest(pts, NP, &a, &b)));
       printf("point (%f,%f) and (%f,%f)\n", a->x, a->y, b->x, b->y);

       return 0;

}</lang> --Ledrug 21:34, 17 June 2011 (UTC)