/**
* \brief Remove a variable from the current parameters and remove it's scaling and prior values
*
* This function will clear out a variable from the \c currentParams and also remove it's scale
* factors and prior ranges.
*
* \param runState [in] The analysis information structure
* \param ifo [in] The IFO data structure
* \param var [in] The variable to remove
*
*/
void remove_variable_and_prior( LALInferenceRunState *runState, LALInferenceIFOModel *ifo, const CHAR *var ){
/* remove variable from currentParams */
if( LALInferenceCheckVariable( runState->currentParams, var ) ){
LALInferenceRemoveVariable( runState->currentParams, var );
}
else{
fprintf(stderr, "Error... variable %s cannot be removed as it does not exist!\n", var);
exit(3);
}
/* remove variable scale parameters from data */
LALInferenceIFOModel *ifotemp = ifo;
while ( ifotemp ){
CHAR *varscale = XLALStringDuplicate( var );
varscale = XLALStringAppend( varscale, "_scale" );
if( LALInferenceCheckVariable( ifotemp->params, varscale ) ){
LALInferenceRemoveVariable( ifotemp->params, varscale );
}
else{
fprintf(stderr, "Error... variable %s cannot be removed as it does not exist!\n", varscale);
exit(3);
}
varscale = XLALStringAppend( varscale, "_min" );
if( LALInferenceCheckVariable( ifotemp->params, varscale ) ){
LALInferenceRemoveVariable( ifotemp->params, varscale );
}
else{
fprintf(stderr, "Error... variable %s cannot be removed as it does not exist!\n", varscale);
exit(3);
}
ifotemp = ifotemp->next;
}
/* remove prior */
if ( LALInferenceCheckGaussianPrior( runState->priorArgs, var ) ){
LALInferenceRemoveGaussianPrior( runState->priorArgs, var );
}
else { LALInferenceRemoveMinMaxPrior( runState->priorArgs, var ); }
}
/**
* \brief Add a variable, checking first if it already exists and is of type \c LALINFERENCE_PARAM_FIXED
* and if so removing it before re-adding it
*
* This function is for use as an alternative to \c LALInferenceAddVariable, which does not
* allow \c LALINFERENCE_PARAM_FIXED variables to be changed. If the variable already exists
* and is of type \c LALINFERENCE_PARAM_FIXED, then it will first be removed and then re-added
*/
void check_and_add_fixed_variable( LALInferenceVariables *vars, const char *name, void *value, LALInferenceVariableType type ){
if ( LALInferenceCheckVariable( vars, name ) ){
if ( LALInferenceGetVariableVaryType( vars, name ) == LALINFERENCE_PARAM_FIXED ) { LALInferenceRemoveVariable( vars, name ); }
}
LALInferenceAddVariable( vars, name, value, type, LALINFERENCE_PARAM_FIXED );
}
int LALInferenceDrawFromPriorTest(void)
{
TEST_HEADER();
int errnum;
const char *name;
gsl_rng *rng = gsl_rng_alloc(gsl_rng_default);
gsl_rng_set(rng, 0);
LALInferenceVariables *output = XLALCalloc(1, sizeof(LALInferenceVariables));
LALInferenceVariables *priorArgs = XLALCalloc(1, sizeof(LALInferenceVariables));
// Null reference checks.
int outcome = 1;
XLAL_TRY(LALInferenceDrawFromPrior(NULL, priorArgs, rng), errnum);
outcome &= errnum == XLAL_EFAULT;
XLAL_TRY(LALInferenceDrawFromPrior(output, NULL, rng), errnum);
outcome &= errnum == XLAL_EFAULT;
XLAL_TRY(LALInferenceDrawFromPrior(output, priorArgs, NULL), errnum);
outcome &= errnum == XLAL_EFAULT;
if (!outcome)
TEST_FAIL("Null reference check failed.");
int i;
const char *varyName=NULL;
char caseTag[VARNAME_MAX];
LALInferenceVariableType type = LALINFERENCE_REAL8_t;
LALInferenceParamVaryType vary=-1;
for (i = 0; i < 2; i++)
{
switch (i)
{
case 0:
vary = LALINFERENCE_PARAM_LINEAR;
varyName = "linear";
break;
case 1:
vary = LALINFERENCE_PARAM_CIRCULAR;
varyName = "circular";
break;
}
sprintf(caseTag, "[%s] ", varyName);
// Try and generate some normally distributed variables for various mu and sigma.
REAL8 gaussian = 0;
REAL8 mu;
REAL8 sigma;
name = "gaussian";
LALInferenceAddVariable(output, name, &gaussian, type, vary);
// Zero standard deviation; should always equal mean.
mu = -50;
sigma = 0;
LALInferenceRemoveGaussianPrior(priorArgs, name);
LALInferenceAddGaussianPrior(priorArgs, name, &mu, &sigma, type);
XLAL_TRY(LALInferenceDrawFromPrior(output, priorArgs, rng), errnum);
gaussian = *(REAL8*)LALInferenceGetVariable(output, name);
if (errnum != XLAL_SUCCESS)
{
TEST_FAIL("%sFailed to generate Gaussian variable; XLAL error: %s.", caseTag, XLALErrorString(errnum));
}
else if (!compareFloats(gaussian, mu, EPSILON))
{
TEST_FAIL("%sGaussian variable with zero standard deviation did not match the mean; X = %f, mu = %f.", caseTag, gaussian, mu);
}
LALInferenceRemoveVariable(output, name);
LALInferenceRemoveGaussianPrior(priorArgs, name);
// Try a uniform variable!
REAL8 uniform = 0;
REAL8 min;
REAL8 max;
name = "uniform";
LALInferenceAddVariable(output, name, &uniform, type, vary);
min = -1;
max = 1;
LALInferenceRemoveMinMaxPrior(priorArgs, name);
LALInferenceAddMinMaxPrior(priorArgs, name, &min, &max, type);
XLAL_TRY(LALInferenceDrawFromPrior(output, priorArgs, rng), errnum);
if (errnum != XLAL_SUCCESS)
TEST_FAIL("%sFailed to generate uniform variable; XLAL error: %s.", caseTag, XLALErrorString(errnum));
LALInferenceRemoveVariable(output, name);
LALInferenceRemoveMinMaxPrior(priorArgs, name);
// Try a correlated variable!
REAL8 correlated = 0;
UINT4 idx = 0;
name = "correlated";
gsl_matrix *covariance = gsl_matrix_calloc(3, 3);
LALInferenceAddVariable(output, name, &correlated, type, vary);
LALInferenceRemoveCorrelatedPrior(priorArgs);
// See what happens when we try to add a non-positive-definite covariance matrix
gsl_matrix_set(covariance, 0, 0, -1);
XLAL_TRY(LALInferenceAddCorrelatedPrior(priorArgs, name, &covariance, &mu, &sigma, &idx), errnum);
if (errnum == XLAL_SUCCESS)
TEST_FAIL("%sNon-positive-definite covariance matrix was not rejected.", caseTag);
LALInferenceRemoveCorrelatedPrior(priorArgs);
// Now try a positive-semi-definite matrix; this should be accepted
covariance = gsl_matrix_calloc(3, 3);
gsl_matrix_set(covariance, 0, 0, 1);
XLAL_TRY(LALInferenceAddCorrelatedPrior(priorArgs, name, &covariance, &mu, &sigma, &idx), errnum);
if (errnum != XLAL_SUCCESS)
TEST_FAIL("%sCould not add semi-positive-definite covariance matrix.", caseTag);
LALInferenceRemoveCorrelatedPrior(priorArgs);
// Try a legitimate positive-definite covariance matrix.
covariance = gsl_matrix_calloc(3, 3);
gsl_matrix_set(covariance, 0, 0, 2);
gsl_matrix_set(covariance, 0, 1, 1);
gsl_matrix_set(covariance, 0, 2, 0);
gsl_matrix_set(covariance, 1, 0, 1);
gsl_matrix_set(covariance, 1, 1, 5);
gsl_matrix_set(covariance, 1, 2, 1);
gsl_matrix_set(covariance, 2, 0, 0);
gsl_matrix_set(covariance, 2, 1, 1);
gsl_matrix_set(covariance, 2, 2, 1);
XLAL_TRY(LALInferenceAddCorrelatedPrior(priorArgs, name, &covariance, &mu, &sigma, &idx), errnum);
if (errnum != XLAL_SUCCESS)
TEST_FAIL("%sCould not add correlated prior.", caseTag);
XLAL_TRY(LALInferenceDrawFromPrior(output, priorArgs, rng), errnum);
if (errnum != XLAL_SUCCESS)
TEST_FAIL("%sCould not generate correlated variable from positive-definite matrix; XLAL error: %s.", caseTag, XLALErrorString(errnum));
LALInferenceRemoveVariable(output, name);
LALInferenceRemoveCorrelatedPrior(priorArgs);
//gsl_matrix_free(covariance);
LALInferenceRemoveVariable(output, "gaussian");
LALInferenceRemoveVariable(output, "uniform");
LALInferenceRemoveVariable(output, "correlated");
}
XLALFree(output);
XLALFree(priorArgs);
TEST_FOOTER();
}