/* * statistics.c * * Structure-factor statistics * * (c) 2006-2010 Thomas White * * Part of CrystFEL - crystallography with a FEL * */ #ifdef HAVE_CONFIG_H #include #endif #include #include #include #include #include #include "statistics.h" #include "utils.h" struct r_params { const double *ref1; const double *ref2; ReflItemList *items; /* Which reflections to use */ int fom; /* Which FoM to use (see the enum just below) */ }; enum { R_1_ZERO, R_1_IGNORE, R_2, R_1_I, R_DIFF_ZERO, R_DIFF_IGNORE, }; /* Return the least squares optimal scaling factor when comparing intensities. * ref1,ref2 are the two intensity lists to compare. "items" is a ReflItemList * containing the reflections which should be taken into account. */ double stat_scale_intensity(const double *ref1, const double *ref2, ReflItemList *items) { double top = 0.0; double bot = 0.0; int i; for ( i=0; ih; k = it->k; l = it->l; i1 = lookup_intensity(ref1, h, k, l); i2 = lookup_intensity(ref2, h, k, l); top += i1 * i2; bot += i2 * i2; } return top/bot; } /* Return the least squares optimal scaling factor when comparing the square * roots of the intensities (i.e. one approximation to the structure factor * moduli). * ref1,ref2 are the two intensity lists to compare (they contain intensities, * not square rooted intensities). "items" is a ReflItemList containing the * reflections which should be taken into account. */ static double stat_scale_sqrti(const double *ref1, const double *ref2, ReflItemList *items) { double top = 0.0; double bot = 0.0; int i; for ( i=0; ih; k = it->k; l = it->l; i1 = lookup_intensity(ref1, h, k, l); if ( i1 < 0.0 ) continue; f1 = sqrt(i1); i2 = lookup_intensity(ref2, h, k, l); if ( i2 < 0.0 ) continue; f2 = sqrt(i2); top += f1 * f2; bot += f2 * f2; } return top/bot; } static double internal_r1_ignorenegs(const double *ref1, const double *ref2, ReflItemList *items, double scale) { double top = 0.0; double bot = 0.0; int i; for ( i=0; ih; k = it->k; l = it->l; i1 = lookup_intensity(ref1, h, k, l); if ( i1 < 0.0 ) continue; f1 = sqrt(i1); i2 = lookup_intensity(ref2, h, k, l); if ( i2 < 0.0 ) continue; f2 = sqrt(i2); f2 *= scale; top += fabs(f1 - f2); bot += f1; } return top/bot; } static double internal_r1_negstozero(const double *ref1, const double *ref2, ReflItemList *items, double scale) { double top = 0.0; double bot = 0.0; int i; for ( i=0; ih; k = it->k; l = it->l; i1 = lookup_intensity(ref1, h, k, l); f1 = i1 > 0.0 ? sqrt(i1) : 0.0; i2 = lookup_intensity(ref2, h, k, l); f2 = i2 > 0.0 ? sqrt(i2) : 0.0; f2 *= scale; top += fabs(f1 - f2); bot += f1; } return top/bot; } static double internal_r2(const double *ref1, const double *ref2, ReflItemList *items, double scale) { double top = 0.0; double bot = 0.0; int i; for ( i=0; ih; k = it->k; l = it->l; i1 = lookup_intensity(ref1, h, k, l); i2 = scale * lookup_intensity(ref2, h, k, l); top += pow(i1 - i2, 2.0); bot += pow(i1, 2.0); } return sqrt(top/bot); } static double internal_r_i(const double *ref1, const double *ref2, ReflItemList *items, double scale) { double top = 0.0; double bot = 0.0; int i; for ( i=0; ih; k = it->k; l = it->l; i1 = lookup_intensity(ref1, h, k, l); i2 = scale * lookup_intensity(ref2, h, k, l); top += fabs(i1-i2); bot += fabs(i1); } return top/bot; } static double internal_rdiff_negstozero(const double *ref1, const double *ref2, ReflItemList *items, double scale) { double top = 0.0; double bot = 0.0; int i; for ( i=0; ih; k = it->k; l = it->l; i1 = lookup_intensity(ref1, h, k, l); f1 = i1 > 0.0 ? sqrt(i1) : 0.0; i2 = lookup_intensity(ref2, h, k, l); f2 = i2 > 0.0 ? sqrt(i2) : 0.0; f2 *= scale; top += fabs(f1 - f2); bot += f1 + f2; } return 2.0*top/bot; } static double internal_rdiff_ignorenegs(const double *ref1, const double *ref2, ReflItemList *items, double scale) { double top = 0.0; double bot = 0.0; int i; for ( i=0; ih; k = it->k; l = it->l; i1 = lookup_intensity(ref1, h, k, l); if ( i1 < 0.0 ) continue; f1 = sqrt(i1); i2 = lookup_intensity(ref2, h, k, l); if ( i2 < 0.0 ) continue; f2 = sqrt(i2); f2 *= scale; top += fabs(f1 - f2); bot += f1 + f2; } return 2.0*top/bot; } static double calc_r(double scale, void *params) { struct r_params *rp = params; switch ( rp->fom) { case R_1_ZERO : return internal_r1_negstozero(rp->ref1, rp->ref2, rp->items, scale); case R_1_IGNORE : return internal_r1_ignorenegs(rp->ref1, rp->ref2, rp->items, scale); case R_2 : return internal_r2(rp->ref1, rp->ref2, rp->items, scale); case R_1_I : return internal_r_i(rp->ref1, rp->ref2, rp->items, scale); case R_DIFF_ZERO : return internal_rdiff_negstozero(rp->ref1, rp->ref2, rp->items, scale); case R_DIFF_IGNORE : return internal_rdiff_ignorenegs(rp->ref1, rp->ref2, rp->items, scale); } ERROR("No such FoM!\n"); abort(); } static double r_minimised(const double *ref1, const double *ref2, ReflItemList *items, double *scalep, int fom) { gsl_function F; gsl_min_fminimizer *s; int status; double scale = 1.0; struct r_params rp; int iter = 0; rp.ref1 = ref1; rp.ref2 = ref2; rp.items = items; rp.fom = fom; F.function = &calc_r; F.params = &rp; s = gsl_min_fminimizer_alloc(gsl_min_fminimizer_brent); /* Initial guess */ switch ( fom ) { case R_1_ZERO : case R_1_IGNORE : case R_DIFF_ZERO : case R_DIFF_IGNORE : scale = stat_scale_sqrti(ref1, ref2, items); break; case R_2 : case R_1_I : scale = stat_scale_intensity(ref1, ref2, items); break; } //STATUS("Initial scale factor estimate: %5.2e\n", scale); /* Probably within an order of magnitude either side */ gsl_min_fminimizer_set(s, &F, scale, scale/10.0, scale*10.0); do { double lo, up; /* Iterate */ gsl_min_fminimizer_iterate(s); /* Get the current estimate */ scale = gsl_min_fminimizer_x_minimum(s); lo = gsl_min_fminimizer_x_lower(s); up = gsl_min_fminimizer_x_upper(s); /* Check for convergence */ status = gsl_min_test_interval(lo, up, 0.001, 0.0); iter++; } while ( status == GSL_CONTINUE ); if (status != GSL_SUCCESS) { ERROR("Scale factor minimisation failed.\n"); } gsl_min_fminimizer_free(s); //STATUS("Final scale factor: %5.2e\n", scale); *scalep = scale; return calc_r(scale, &rp); } double stat_r1_ignore(const double *ref1, const double *ref2, ReflItemList *items, double *scalep) { return r_minimised(ref1, ref2, items, scalep, R_1_IGNORE); } double stat_r1_zero(const double *ref1, const double *ref2, ReflItemList *items, double *scalep) { return r_minimised(ref1, ref2, items, scalep, R_1_ZERO); } double stat_r2(const double *ref1, const double *ref2, ReflItemList *items, double *scalep) { return r_minimised(ref1, ref2, items, scalep, R_2); } double stat_r1_i(const double *ref1, const double *ref2, ReflItemList *items, double *scalep) { return r_minimised(ref1, ref2, items, scalep, R_1_I); } double stat_rdiff_zero(const double *ref1, const double *ref2, ReflItemList *items, double *scalep) { return r_minimised(ref1, ref2, items, scalep, R_DIFF_ZERO); } double stat_rdiff_ignore(const double *ref1, const double *ref2, ReflItemList *items, double *scalep) { return r_minimised(ref1, ref2, items, scalep, R_DIFF_IGNORE); } double stat_pearson_i(const double *ref1, const double *ref2, ReflItemList *items) { double *vec1, *vec2; int i = 0; int ni = num_items(items); double val; vec1 = malloc(ni*sizeof(double)); vec2 = malloc(ni*sizeof(double)); for ( i=0; ih; k = it->k; l = it->l; i1 = lookup_intensity(ref1, h, k, l); i2 = lookup_intensity(ref2, h, k, l); vec1[i] = i1; vec2[i] = i2; printf("%f %f\n", i1, i2); } val = gsl_stats_correlation(vec1, 1, vec2, 1, i); free(vec1); free(vec2); return val; } double stat_pearson_f_ignore(const double *ref1, const double *ref2, ReflItemList *items) { double *vec1, *vec2; int i = 0; int ni = num_items(items); double val; vec1 = malloc(ni*sizeof(double)); vec2 = malloc(ni*sizeof(double)); for ( i=0; ih; k = it->k; l = it->l; i1 = lookup_intensity(ref1, h, k, l); if ( i1 < 0.0 ) continue; f1 = sqrt(i1); i2 = lookup_intensity(ref2, h, k, l); if ( i2 < 0.0 ) continue; f2 = sqrt(i2); vec1[i] = f1; vec2[i] = f2; } val = gsl_stats_correlation(vec1, 1, vec2, 1, i); free(vec1); free(vec2); return val; } double stat_pearson_f_zero(const double *ref1, const double *ref2, ReflItemList *items) { double *vec1, *vec2; int i = 0; int ni = num_items(items); double val; vec1 = malloc(ni*sizeof(double)); vec2 = malloc(ni*sizeof(double)); for ( i=0; ih; k = it->k; l = it->l; i1 = lookup_intensity(ref1, h, k, l); f1 = i1 > 0.0 ? sqrt(i1) : 0.0; i2 = lookup_intensity(ref2, h, k, l); f2 = i2 > 0.0 ? sqrt(i2) : 0.0; vec1[i] = f1; vec2[i] = f2; } val = gsl_stats_correlation(vec1, 1, vec2, 1, i); free(vec1); free(vec2); return val; }