src/suave/Sample.c


1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171

/*
	Sample.c
		the sampling step of Suave
		this file is part of Suave
		last modified 12 Aug 11 th
*/


typedef struct {
  real sum, sqsum;
  real weight, weightsum, avg, avgsum;
  real guess, chisum, chisqsum;
} Cumulants;

/*********************************************************************/

static void Sample(This *t, cnumber nnew, void *voidregion,
  real *lastw, real *lastx, real *lastf)
{
  TYPEDEFREGION;

  Region *const region = (Region *)voidregion;
  count comp, dim, df;
  number n;
  Cumulants cumul[NCOMP];
  char **ss, *s;
  ccount chars = 128*(region->div + 1);

  creal jacobian = 1/ldexp((real)nnew, region->div);
  real *w = lastw, *f = lastx;
  bin_t *bin = (bin_t *)(lastf + nnew*t->ncomp);

  for( n = nnew; n; --n ) {
    real weight = jacobian;

    t->rng.getrandom(t, f);

    for( dim = 0; dim < t->ndim; ++dim ) {
      cBounds *b = &region->bounds[dim];
      creal pos = *f*NBINS;
      ccount ipos = (count)pos;
      creal prev = (ipos == 0) ? 0 : b->grid[ipos - 1];
      creal diff = b->grid[ipos] - prev;
      *f++ = b->lower + (prev + (pos - ipos)*diff)*(b->upper - b->lower);
      *bin++ = ipos;
      weight *= diff*NBINS;
    }

    *w++ = weight;
  }

  DoSample(t, nnew, lastx, lastf, lastw, region->div + 1);

  w[-1] = -w[-1];
  lastw = w;
  w = region->w;
  region->n = lastw - w;

  if( VERBOSE > 2 ) {
    char *p0;
    MemAlloc(ss, t->ndim*64 + t->ncomp*(sizeof(char *) + chars));
    s = (char *)(ss + t->ncomp);
    p0 = s + t->ndim*64;
    for( comp = 0; comp < t->ncomp; ++comp ) {
      ss[comp] = p0;
      p0 += chars;
    }
  }

  Zap(cumul);
  df = n = 0;

  while( w < lastw ) {
    cbool final = (*w < 0);
    creal weight = fabs(*w++);
    ++n;

    for( comp = 0; comp < t->ncomp; ++comp ) {
      Cumulants *c = &cumul[comp];

      creal wfun = weight*(*f++);
      c->sum += wfun;
      c->sqsum += Sq(wfun);

      if( final ) {
        if( n > 1 ) {
          real w = Weight(c->sum, c->sqsum, n);
          c->weightsum += c->weight = w;
          c->avgsum += c->avg = w*c->sum;

          if( VERBOSE > 2 ) {
            creal sig = sqrt(1/w);
            ss[comp] += (df == 0) ?
              sprintf(ss[comp], "\n[" COUNT "] "
                REAL " +- " REAL " (" NUMBER ")", comp + 1,
                c->sum, sig, n) :
              sprintf(ss[comp], "\n    "
                REAL " +- " REAL " (" NUMBER ")",
                c->sum, sig, n);
          }

          if( df == 0 ) c->guess = c->sum;
          else {
            c->chisum += w *= c->sum - c->guess;
            c->chisqsum += w*c->sum;
          }
        }

        c->sum = c->sqsum = 0;
      }
    }

    if( final ) ++df, n = 0;
  }

  region->df = --df;

  for( comp = 0; comp < t->ncomp; ++comp ) {
    Result *r = &region->result[comp];
    Cumulants *c = &cumul[comp];
    creal sigsq = 1/c->weightsum;
    creal avg = sigsq*c->avgsum;

    if( LAST ) {
      r->sigsq = 1/c->weight;
      r->avg = r->sigsq*c->avg;
    }
    else {
      r->sigsq = sigsq;
      r->avg = avg;
    }
    r->err = sqrt(r->sigsq);

    r->chisq = (sigsq < .9*NOTZERO) ? 0 : c->chisqsum - avg*c->chisum;
      /* This catches the special case where the integrand is constant
         over the entire region.  Unless that constant is zero, only the
         first set of samples will have zero variance, and hence weight
         (n - 1) 1e30 (see above).  All other sets have been sampled
         from a non-constant weight function and therefore inevitably
         show some variance.  This is an artificial effect, brought about
         by the fact that the constancy of the integrand in the region is
         seen only in this subdivision, and can degrade the chi-square
         score quite a bit.  If the constancy was determined from more
         than two samples (hence .9*NOTZERO), the chi-squares from the
         other sets are removed here. */
  }

  if( VERBOSE > 2 ) {
    char *p = s;
    char *p0 = p + t->ndim*64;

    for( dim = 0; dim < t->ndim; ++dim ) {
      cBounds *b = &region->bounds[dim];
      p += sprintf(p, 
        (dim == 0) ? "\nRegion (" REALF ") - (" REALF ")" :
                     "\n       (" REALF ") - (" REALF ")",
        b->lower, b->upper);
    }

    for( comp = 0; comp < t->ncomp; ++comp ) {
      cResult *r = &region->result[comp];
      p += sprintf(p, "%s  \tchisq " REAL " (" COUNT " df)",
        p0, r->chisq, df);
      p0 += chars;
    }

    Print(s);
    free(ss);
  }
}