int main(void)
{
long idum=(-1984);
int i,mwt=1;
float a,abdev,b,chi2,q,siga,sigb;
float *x,*y,*sig;
x=vector(1,NDATA);
y=vector(1,NDATA);
sig=vector(1,NDATA);
for (i=1;i<=NPT;i++) {
x[i]=0.1*i;
y[i] = -2.0*x[i]+1.0+SPREAD*gasdev(&idum);
sig[i]=SPREAD;
}
fit(x,y,NPT,sig,mwt,&a,&b,&siga,&sigb,&chi2,&q);
printf("\nAccording to routine FIT the result is:\n");
printf(" a = %8.4f uncertainty: %8.4f\n",a,siga);
printf(" b = %8.4f uncertainty: %8.4f\n",b,sigb);
printf(" chi-squared: %8.4f for %4d points\n",chi2,NPT);
printf(" goodness-of-fit: %8.4f\n",q);
printf("\nAccording to routine MEDFIT the result is:\n");
medfit(x,y,NPT,&a,&b,&abdev);
printf(" a = %8.4f\n",a);
printf(" b = %8.4f\n",b);
printf(" absolute deviation (per data point): %8.4f\n",abdev);
printf(" (note: gaussian SPREAD is %8.4f)\n",SPREAD);
free_vector(sig,1,NDATA);
free_vector(y,1,NDATA);
free_vector(x,1,NDATA);
return 0;
}
void spectrum_wrapper(control *c, timeseries *ts, double *xdata, double *ydata, double *frequency, double *power)
{
/*
wrapper func - to calculate periodgram for given time-series
also has the options to remove trends from the data and apply
a window.
*/
int i, j, istart;
double intercept = 0., abdev = 0., slope = 0.;
double *ftrx = NULL, *ftix = NULL, *xwk = NULL, *ywk = NULL, scal;
ftrx = allocate_memory_double(ts->lfreq+1, __LINE__);
ftix = allocate_memory_double(ts->lfreq+1, __LINE__);
xwk = allocate_memory_double(ts->nseg+1, __LINE__); /* periodgram power */
ywk = allocate_memory_double(ts->nseg+1, __LINE__); /* corresponding frequencies */
/* scale autospectrum and setup frequency axis */
scal = 2.0 / ((double)ts->n50 * (double)ts->nseg * ts->df * ts->ofac);
ts->factor = scal; /* store for coherency code */
for (i = 1; i <= ts->n50; i++) {
/* copy data of i'th segment into workspace */
istart = (int)((i - 1) * (double)ts->nseg / 2.);
for (j = 1; j <= ts->nseg; j++) {
xwk[j] = xdata[istart + j];
ywk[j] = ydata[istart + j];
}
/* detrend the data */
if (c->DETREND == TRUE) {
rmtrend(c, ts, ywk, xwk);
} else if (c->ROBUSTDETREND == TRUE) {
/*
fitting method that is more sensitive to outliers in
the time-series, I guess the way to go?
*/
medfit(ts->nseg, ywk, xwk, &intercept, &slope, &abdev);
/* Subtract the fitted line from the time-series to obtain a linear detrend */
for (j = 1; j <= ts->nseg; j++) {
ywk[j] -= (slope * xwk[j]) + intercept;
}
}
/* apply window to data */
if (c->WINDOW_TYPE != NO_WINDOW) {
if (c->WINDOW_TYPE == COSINE_TAPER) {
taper_timeseries(c, ts, ywk); /* leads to some bizarre results if used in coherency */
} else {
window(c, ts, ywk, xwk);
}
}
/*
calculate periodgram
- there are two options the original implementation in Scargle paper
or the Press et al method. They give identical results.
*/
if (c->PERIODOGRAM_TYPE == SCARGLE) {
ft_uneven_data(xwk, ywk, ftrx, ftix, ts->nfreq, ts->nseg, ts->lfreq, ts->wz);
/* sum raw spectra */
for (j = 1; j <= ts->nout; j++) {
power[j] += ftrx[j] * ftrx[j] + ftix[j] * ftix[j];
}
} else if (c->PERIODOGRAM_TYPE == PRESS) {
period(xwk, ywk, ftrx, ftix, ts->nseg, ts->df, ts->nout);
/* sum raw spectra */
for (j = 1; j <= ts->nout; j++) {
power[j] += ftrx[j] * ftrx[j] + ftix[j] * ftix[j];
}
}
for (j = 1; j <= ts->nseg; j++) {
xwk[j] = 0.0;
ywk[j] = 0.0;
}
}
/* rescale periodogram */
if (c->PERIODOGRAM_TYPE == SCARGLE) {
for (i = 1; i <= ts->nout; i++) {
power[i] *= scal;
frequency[i] = (i - 1) * ts->df;
}
} else if (c->PERIODOGRAM_TYPE == PRESS) {
for (i = 1; i <= ts->nout; i++) {
power[i] *= scal;
/* press algorithm starts at df, rather than 0.0 */
frequency[i] = i * ts->df;
}
}
/* tidy up */
free_array_double(xwk);
free_array_double(ywk);
free_array_double(ftrx);
free_array_double(ftix);
return;
}