double anova(double *vec, int len, int *xtypes, int numeg)
// anova 1 but f statistic
{
int i, k ;
double y1, top, bot, ftail ;
double *w0, *w1, *popsize, *wmean ;
static int ncall2 = 0 ;
if (numeg >= len) fatalx("bad anova\n") ;
ZALLOC(w0, len, double) ;
ZALLOC(w1, len, double) ;
ZALLOC(wmean, numeg, double) ;
ZALLOC(popsize, numeg, double) ;
y1 = asum(vec, len)/ (double) len ; // mean
vsp(w0, vec, -y1, len) ;
for (i=0; i<len; i++) {
k = xtypes[i] ;
++popsize[k] ;
wmean[k] += w0[i] ;
}
/* debug */
if (numeg == 2) {
++ncall2 ;
for (i=0; i<len; ++i) {
if (ncall2<0) break ;
k = xtypes[i] ;
// printf("yy %4d %4d %12.6f %12.6f\n", i, k, vec[i], w0[i]) ;
}
}
vsp(popsize, popsize, 1.0e-12, numeg) ;
vvd(wmean, wmean, popsize, numeg) ;
vvt(w1, wmean, wmean, numeg) ;
top = vdot(w1, popsize, numeg) ;
for (i=0; i<len ; i++) {
k = xtypes[i] ;
w1[i] = w0[i] - wmean[k] ;
}
bot = asum2(w1, len) / (double) (len-numeg) ;
bot *= (double) (numeg-1) ;
ftail = rtlf(numeg-1, len-numeg, top/bot) ;
free(w0) ;
free(w1) ;
free(popsize) ;
free(wmean) ;
return ftail ;
}
double
anova1 (double *vec, int len, int *xtypes, int numeg)
{
int i, k;
double y1, y2, ylike;
double *w0, *w1, *popsize, *wmean;
ZALLOC(w0, len, double);
ZALLOC(w1, len, double);
ZALLOC(wmean, numeg, double);
ZALLOC(popsize, numeg, double);
y1 = asum (vec, len) / (double) len; // mean
vsp (w0, vec, -y1, len);
for (i = 0; i < len; i++)
{
k = xtypes[i];
++popsize[k];
wmean[k] += w0[i];
}
vsp (popsize, popsize, 1.0e-12, numeg);
vvd (wmean, wmean, popsize, numeg);
for (i = 0; i < len; i++)
{
k = xtypes[i];
w1[i] = w0[i] - wmean[k];
}
y1 = asum2 (w0, len) / (double) len;
y2 = asum2 (w1, len) / (double) len;
ylike = 0.5 * ((double) len) * log (y1 / y2);
free (w0);
free (w1);
free (popsize);
free (wmean);
return ylike;
}
double chitest(double *a, double *p, int n)
/* a is n boxes. Goodness of fit test to p */
{
double *x, *b, *pp ;
double y1=0.0, y2=0.0 ;
int i ;
ZALLOC(pp, n, double) ;
if (p != NULL)
copyarr(p,pp,n) ;
else
vclear(pp, 1.0, n) ;
y1 = asum(pp,n) ;
y2 = asum(a,n) ;
if ( (y1==0.0) || (y2==0.0) ) {
free(pp) ;
return 0.0 ;
}
ZALLOC(x,n,double) ;
ZALLOC(b,n,double) ;
vst (x, pp, y2/y1, n) ; /* expected */
vsp (x, x, .0001, n) ;
vvm (b, a, x, n) ;
vvt (b, b, b, n) ;
vvd (b, b, x, n) ;
y1 = asum(b,n) ;
free(x) ;
free(b) ;
return y1 ;
}
void
dotpops(double *X, char **eglist, int numeg, int *xtypes, int nrows)
{
double *pp, *npp, val, yy ;
int *popsize ;
int i, j, k1, k2 ;
if (fstonly) return ;
ZALLOC(pp, numeg * numeg, double) ;
ZALLOC(npp, numeg * numeg, double) ;
popsize = xpopsize;
ivzero(popsize, numeg) ;
for (i=0; i<nrows; i++) {
k1 = xtypes[i] ;
++popsize[k1] ;
for (j=i+1; j<nrows; j++) {
k2 = xtypes[j] ;
if (k1 < 0) fatalx("bug\n") ;
if (k2 < 0) fatalx("bug\n") ;
if (k1>=numeg) fatalx("bug\n") ;
if (k2>=numeg) fatalx("bug\n") ;
val = X[i*nrows+i] + X[j*nrows+j] - 2.0*X[i*nrows+j] ;
pp[k1*numeg+k2] += val ;
pp[k2*numeg+k1] += val ;
++npp[k1*numeg+k2] ;
++npp[k2*numeg+k1] ;
}
}
vsp(npp, npp, 1.0e-8, numeg*numeg) ;
vvd(pp, pp, npp, numeg*numeg) ;
// and normalize so that mean on diagonal is 1
yy = trace(pp, numeg) / (double) numeg ;
vst(pp, pp, 1.0/yy, numeg*numeg) ;
printf("\n## Average divergence between populations:");
if (numeg<=10) {
printf("\n") ;
printf("%10s", "") ;
for (k1=0; k1<numeg; ++k1) {
printf(" %10s", eglist[k1]) ;
}
printf(" %10s", "popsize") ;
printf("\n") ;
for (k2=0; k2<numeg; ++k2) {
printf("%10s", eglist[k2]) ;
for (k1=0; k1<numeg; ++k1) {
val = pp[k1*numeg+k2] ;
printf(" %10.3f", val) ;
}
printf(" %10d", popsize[k2]) ;
printf("\n") ;
}
}
else { // numeg >= 10
printf("\n") ;
for (k2=0; k2<numeg; ++k2) {
for (k1=k2; k1<numeg; ++k1) {
printf("dotp: %10s", eglist[k2]) ;
printf(" %10s", eglist[k1]) ;
val = pp[k1*numeg+k2] ;
printf(" %10.3f", val) ;
printf(" %10d", popsize[k2]) ;
printf(" %10d", popsize[k1]) ;
printf("\n") ;
}
}
}
printf("\n") ;
printf("\n") ;
fflush(stdout) ;
free(pp) ;
free(npp) ;
}
