Author: bugman
Date: Fri Aug 29 09:16:51 2014
New Revision: 25414
URL: http://svn.gna.org/viewcvs/relax?rev=25414&view=rev
Log:
Speed up of the target_functions.relax_fit C module.
The variances are now precalculated in the setup() function from the errors, so
that the use of the
square() function is minimised. The chi-squared equation, gradient, and
Hessian functions now
accept the variance rather than standard deviation argument and hence the
squaring of errors has
been removed. This avoids a lot of duplicated maths operations.
Modified:
trunk/target_functions/c_chi2.c
trunk/target_functions/c_chi2.h
trunk/target_functions/relax_fit.c
trunk/target_functions/relax_fit.h
Modified: trunk/target_functions/c_chi2.c
URL:
http://svn.gna.org/viewcvs/relax/trunk/target_functions/c_chi2.c?rev=25414&r1=25413&r2=25414&view=diff
==============================================================================
--- trunk/target_functions/c_chi2.c (original)
+++ trunk/target_functions/c_chi2.c Fri Aug 29 09:16:51 2014
@@ -21,7 +21,7 @@
#include "c_chi2.h"
-double chi2(double values[MAX_DATA], double sd[MAX_DATA], double
back_calc[MAX_DATA], int num_times) {
+double chi2(double values[MAX_DATA], double variance[MAX_DATA], double
back_calc[MAX_DATA], int num_times) {
/* Function to calculate the chi-squared value.
The chi-sqared equation
@@ -50,14 +50,14 @@
/* Loop over the time points and sum the chi-squared components. */
for (i = 0; i < num_times; ++i) {
- chi2 = chi2 + square((values[i] - back_calc[i]) / sd[i]);
+ chi2 = chi2 + square(values[i] - back_calc[i]) / variance[i];
}
return chi2;
}
-void dchi2(double dchi2[MAX_PARAMS], double data[MAX_DATA], double
back_calc_vals[MAX_DATA], double back_calc_grad[MAX_PARAMS][MAX_DATA], double
errors[MAX_DATA], int num_points, int num_params) {
+void dchi2(double dchi2[MAX_PARAMS], double data[MAX_DATA], double
back_calc_vals[MAX_DATA], double back_calc_grad[MAX_PARAMS][MAX_DATA], double
variance[MAX_DATA], int num_points, int num_params) {
/* Calculate the full chi-squared gradient.
The chi-squared gradient
@@ -90,8 +90,8 @@
@type back_calc_vals: numpy rank-1 size N array
@param back_calc_grad: The matrix of dyi(theta)/dtheta values.
@type back_calc_grad: numpy rank-2 size MxN array
- @param errors: The vector of sigma_i values.
- @type errors: numpy rank-1 size N array
+ @param variance: The vector of sigma_i values squared.
+ @type variance: numpy rank-1 size N array
@param num_points: The number of data points to sum over.
@type num_points: int
@param num_params: The dimensions of the gradient.
@@ -105,13 +105,13 @@
for (j = 0; j < num_params; ++j) {
dchi2[j] = 0.0;
for (i = 0; i < num_points; ++i) {
- dchi2[j] += -2.0 / square(errors[i]) * (data[i] -
back_calc_vals[i]) * back_calc_grad[j][i];
+ dchi2[j] += -2.0 / variance[i] * (data[i] - back_calc_vals[i]) *
back_calc_grad[j][i];
}
}
}
-void d2chi2(double d2chi2[MAX_PARAMS][MAX_PARAMS], double data[MAX_DATA],
double back_calc_vals[MAX_DATA], double back_calc_grad[MAX_PARAMS][MAX_DATA],
double back_calc_hess[MAX_PARAMS][MAX_PARAMS][MAX_DATA], double
errors[MAX_DATA], int num_points, int num_params) {
+void d2chi2(double d2chi2[MAX_PARAMS][MAX_PARAMS], double data[MAX_DATA],
double back_calc_vals[MAX_DATA], double back_calc_grad[MAX_PARAMS][MAX_DATA],
double back_calc_hess[MAX_PARAMS][MAX_PARAMS][MAX_DATA], double
variance[MAX_DATA], int num_points, int num_params) {
/* Calculate the full chi-squared Hessian.
The chi-squared Hessian
@@ -148,8 +148,8 @@
@type back_calc_grad: numpy rank-2 size MxN array
@param back_calc_hess: The matrix of d2yi(theta)/dtheta.dtheta values.
@type back_calc_hess: numpy rank-3 size MxMxN array
- @param errors: The vector of sigma_i values.
- @type errors: numpy rank-1 size N array
+ @param variance: The vector of sigma_i values squared.
+ @type variance: numpy rank-1 size N array
@param num_points: The number of data points to sum over.
@type num_points: int
@param num_params: The dimensions of the Hessian.
@@ -164,7 +164,7 @@
for (k = 0; k < num_params; ++k) {
d2chi2[j][k] = 0.0;
for (i = 0; i < num_points; ++i) {
- d2chi2[j][k] += 2.0 / square(errors[i]) *
(back_calc_grad[j][i] * back_calc_grad[k][i] - (data[i] - back_calc_vals[i]) *
back_calc_hess[j][k][i]);
+ d2chi2[j][k] += 2.0 / variance[i] * (back_calc_grad[j][i] *
back_calc_grad[k][i] - (data[i] - back_calc_vals[i]) * back_calc_hess[j][k][i]);
}
}
}
Modified: trunk/target_functions/c_chi2.h
URL:
http://svn.gna.org/viewcvs/relax/trunk/target_functions/c_chi2.h?rev=25414&r1=25413&r2=25414&view=diff
==============================================================================
--- trunk/target_functions/c_chi2.h (original)
+++ trunk/target_functions/c_chi2.h Fri Aug 29 09:16:51 2014
@@ -26,9 +26,9 @@
#define RELAX_C_CHI2
/* Define all of the functions. */
-double chi2(double values[MAX_DATA], double sd[MAX_DATA], double
back_calc[MAX_DATA], int num_times);
-void dchi2(double dchi2[MAX_PARAMS], double data[MAX_DATA], double
back_calc_vals[MAX_DATA], double back_calc_grad[MAX_PARAMS][MAX_DATA], double
errors[MAX_DATA], int num_times, int M);
-void d2chi2(double d2chi2[MAX_PARAMS][MAX_PARAMS], double data[MAX_DATA],
double back_calc_vals[MAX_DATA], double back_calc_grad[MAX_PARAMS][MAX_DATA],
double back_calc_hess[MAX_PARAMS][MAX_PARAMS][MAX_DATA], double
errors[MAX_DATA], int num_times, int M);
+double chi2(double values[MAX_DATA], double variance[MAX_DATA], double
back_calc[MAX_DATA], int num_times);
+void dchi2(double dchi2[MAX_PARAMS], double data[MAX_DATA], double
back_calc_vals[MAX_DATA], double back_calc_grad[MAX_PARAMS][MAX_DATA], double
variance[MAX_DATA], int num_times, int M);
+void d2chi2(double d2chi2[MAX_PARAMS][MAX_PARAMS], double data[MAX_DATA],
double back_calc_vals[MAX_DATA], double back_calc_grad[MAX_PARAMS][MAX_DATA],
double back_calc_hess[MAX_PARAMS][MAX_PARAMS][MAX_DATA], double
variance[MAX_DATA], int num_times, int M);
/* Define the function for calculating the square of a number. */
#define square(x) ((x)*(x))
Modified: trunk/target_functions/relax_fit.c
URL:
http://svn.gna.org/viewcvs/relax/trunk/target_functions/relax_fit.c?rev=25414&r1=25413&r2=25414&view=diff
==============================================================================
--- trunk/target_functions/relax_fit.c (original)
+++ trunk/target_functions/relax_fit.c Fri Aug 29 09:16:51 2014
@@ -66,6 +66,9 @@
sd[i] = PyFloat_AsDouble(element);
Py_CLEAR(element);
+ /* Convert the errors to variances to avoid duplicated maths
operations for faster calculations. */
+ variance[i] = square(sd[i]);
+
/* The relax_times argument element. */
element = PySequence_GetItem(relax_times_arg, i);
relax_times[i] = PyFloat_AsDouble(element);
@@ -120,7 +123,7 @@
exponential(params[index_I0], params[index_R], relax_times, back_calc,
num_times);
/* Calculate and return the chi-squared value. */
- return PyFloat_FromDouble(chi2(values, sd, back_calc, num_times));
+ return PyFloat_FromDouble(chi2(values, variance, back_calc, num_times));
}
@@ -150,7 +153,7 @@
exponential_dI0(params[index_I0], params[index_R], index_I0, relax_times,
back_calc_grad, num_times);
/* The chi-squared gradient. */
- dchi2(dchi2_vals, values, back_calc, back_calc_grad, sd, num_times,
num_params);
+ dchi2(dchi2_vals, values, back_calc, back_calc_grad, variance, num_times,
num_params);
/* Convert to a Python list, and scale the values. */
list = PyList_New(0);
@@ -194,7 +197,7 @@
exponential_dR_dI0(params[index_I0], params[index_R], index_R, index_I0,
relax_times, back_calc_hess, num_times);
/* The chi-squared Hessian. */
- d2chi2(d2chi2_vals, values, back_calc, back_calc_grad, back_calc_hess, sd,
num_times, num_params);
+ d2chi2(d2chi2_vals, values, back_calc, back_calc_grad, back_calc_hess,
variance, num_times, num_params);
/* Convert to a Python list, and scale the values. */
list = PyList_New(0);
@@ -276,7 +279,7 @@
/* Assemble the chi-squared Jacobian. */
for (j = 0; j < num_params; ++j) {
for (i = 0; i < num_times; ++i) {
- jacobian_matrix[j][i] = -2.0 / square(sd[i]) * (values[i] -
back_calc[i]) * back_calc_grad[j][i];
+ jacobian_matrix[j][i] = -2.0 / variance[i] * (values[i] -
back_calc[i]) * back_calc_grad[j][i];
}
}
Modified: trunk/target_functions/relax_fit.h
URL:
http://svn.gna.org/viewcvs/relax/trunk/target_functions/relax_fit.h?rev=25414&r1=25413&r2=25414&view=diff
==============================================================================
--- trunk/target_functions/relax_fit.h (original)
+++ trunk/target_functions/relax_fit.h Fri Aug 29 09:16:51 2014
@@ -53,3 +53,4 @@
static double sd[MAX_DATA];
static double relax_times[MAX_DATA];
static double scaling_matrix[MAX_PARAMS];
+static double variance[MAX_DATA];
_______________________________________________
relax (http://www.nmr-relax.com)
This is the relax-commits mailing list
[email protected]
To unsubscribe from this list, get a password
reminder, or change your subscription options,
visit the list information page at
https://mail.gna.org/listinfo/relax-commits
