/* The gradient of the NLL w.r.t. each free parameter in p.
*/
static void
hyperexp_complete_gradient(double *p, int np, void *dptr, double *dp)
{
struct hyperexp_data *data = (struct hyperexp_data *) dptr;
ESL_HYPEREXP *h = data->h;
double pdf;
int i,k;
int pidx;
hyperexp_unpack_paramvector(p, np, h);
esl_vec_DSet(dp, np, 0.);
for (i = 0; i < data->n; i++)
{
/* FIXME: I think the calculation below may need to be done
* in log space, to avoid underflow errors; see complete_binned_gradient()
*/
/* Precalculate q_k PDF_k(x) terms, and their sum */
for (k = 0; k < h->K; k++)
h->wrk[k] = h->q[k] * esl_exp_pdf(data->x[i], h->mu, h->lambda[k]);
pdf = esl_vec_DSum(h->wrk, h->K);
pidx = 0;
if (! h->fixmix) {
for (k = 1; k < h->K; k++) /* generic d/dQ solution for mixture models */
dp[pidx++] -= h->wrk[k]/pdf - h->q[k];
}
for (k = 0; k < h->K; k++)
if (! h->fixlambda[k])
dp[pidx++] -= (1.-h->lambda[k]*(data->x[i]-h->mu))*h->wrk[k]/pdf; /* d/dw */
}
}
/* Function: esl_hxp_pdf()
*
* Purpose: Returns the probability density function $P(X=x)$ for
* quantile <x>, given hyperexponential parameters <h>.
*/
double
esl_hxp_pdf(double x, ESL_HYPEREXP *h)
{
double pdf = 0.;
int k;
if (x < h->mu) return 0.;
for (k = 0; k < h->K; k++)
pdf += h->q[k] * esl_exp_pdf(x, h->mu, h->lambda[k]);
return pdf;
}
/* Function: esl_exp_generic_pdf()
* Incept: SRE, Thu Aug 25 07:58:34 2005 [St. Louis]
*
* Purpose: Generic-API version of PDF.
*/
double
esl_exp_generic_pdf(double x, void *params)
{
double *p = (double *) params;
return esl_exp_pdf(x, p[0], p[1]);
}
