write the fit results to an easy-to-read/parse yaml file

This patch adds routines for writing a comprehensive set of fit results (i.e., parameter values, parabolic errors, asymmetric errors, covariances, correlation coefficients, etc.) for an individual `.msr` file to an easy-to-read/parse `.yaml` file.

The main motivation for the code addition is to provide users with easy access to the fit's covariance matrix without the need for "extra" manual effort (e.g., parsing the contents of `MINUIT2.OUTPUT` or `MINUIT2.root`). The other fit quantities are also included for completeness.

Reading/accessing the output is trivial using, for example, the PyYAML Python library (https://github.com/yaml/pyyaml):

```python
import yaml

with open("2125_tf_histo.yaml", "r") as fh:
	results = yaml.load(fh, Loader=yaml.SafeLoader)

cov = results["covariance"]["Field_1"]["Sigma_1"]
```

Note: the naming conventions chosen for the blocks in the `.yaml` output closely follow those used by the iminuit Python library (https://github.com/scikit-hep/iminuit).
This commit is contained in:
Ryan M. L. McFadden 2024-06-17 12:27:33 -03:00
parent b71dce9291
commit 833171c712

View File

@ -39,9 +39,10 @@
#include <iomanip> #include <iomanip>
#include <fstream> #include <fstream>
#include <limits> #include <limits>
#include <cmath> #include <cmath>
#include <boost/variant/variant.hpp>
#include <sys/time.h> #include <sys/time.h>
#include "Minuit2/FunctionMinimum.h" #include "Minuit2/FunctionMinimum.h"
@ -2473,6 +2474,121 @@ Bool_t PFitter::ExecuteSave(Bool_t firstSave)
ccorr->Write("ccorr", TObject::kOverwrite, sizeof(ccorr)); ccorr->Write("ccorr", TObject::kOverwrite, sizeof(ccorr));
hcorr->Write("hcorr", TObject::kOverwrite, sizeof(hcorr)); hcorr->Write("hcorr", TObject::kOverwrite, sizeof(hcorr));
ff.Close(); ff.Close();
// write the fit results to an easy-to-read/parse yaml file
// note: the block names follow those used by Python library iminuit
// https://github.com/scikit-hep/iminuit
// dynamically name the yaml output file
// https://stackoverflow.com/a/25389052
std::string yaml_filename(fRunInfo->GetFileName().Data());
const std::string msr_ext(".msr");
yaml_filename.replace(yaml_filename.find(msr_ext), msr_ext.length(),
".yaml");
// define yaml's 2-space indentation
const std::string yaml_indent(" ");
// open the yaml file for writing
std::ofstream yaml_file(yaml_filename);
// number formatting of the output
yaml_file << std::scientific << std::setprecision(16);
// write the parameter values
yaml_file << "values:\n";
for (unsigned int i = 0; i < fParams.size(); ++i) {
yaml_file << yaml_indent << fParams[i].fName.Data() << ": "
<< fParams[i].fValue << "\n";
}
// write the parabolic errors
yaml_file << "errors:\n";
for (unsigned int i = 0; i < fParams.size(); ++i) {
yaml_file << yaml_indent << fParams[i].fName.Data() << ": "
<< fMnUserParams.Error(i) << "\n";
}
// write the minos errors
yaml_file << "mnerrors:\n";
for (unsigned int i = 0; i < fParams.size(); ++i) {
// use boost's implementation of a variant, which can be streamed by
// default - see: https://stackoverflow.com/q/47168477
boost::variant<double, std::string> positive_error, negative_error;
if (fParams[i].fPosErrorPresent) {
positive_error = fParams[i].fPosError;
negative_error = fParams[i].fStep;
} else {
positive_error = "null";
negative_error = "null";
}
yaml_file << yaml_indent << fParams[i].fName.Data() << ":\n";
yaml_file << yaml_indent << yaml_indent
<< "positive: " << positive_error << "\n";
yaml_file << yaml_indent << yaml_indent
<< "negative: " << negative_error << "\n";
}
// write the parameter limits
yaml_file << "limits:\n";
for (unsigned int i = 0; i < fParams.size(); ++i) {
// use boost's implementation of a variant, which can be streamed by
// default - see: https://stackoverflow.com/q/47168477
boost::variant<double, std::string> upper_limit, lower_limit;
if (fParams[i].fLowerBoundaryPresent) {
lower_limit = fParams[i].fLowerBoundary;
} else {
lower_limit = "null";
}
if (fParams[i].fUpperBoundaryPresent) {
upper_limit = fParams[i].fUpperBoundary;
} else {
upper_limit = "null";
}
yaml_file << yaml_indent << fParams[i].fName.Data() << ":\n";
yaml_file << yaml_indent << yaml_indent << "lower: " << lower_limit
<< "\n";
yaml_file << yaml_indent << yaml_indent << "upper: " << upper_limit
<< "\n";
}
// write if the parameter is fixed
yaml_file << "fixed:\n";
for (unsigned int i = 0; i < fParams.size(); ++i) {
std::string is_fixed = fParams[i].fStep == 0.0 ? "true" : "false";
yaml_file << yaml_indent << fParams[i].fName.Data() << ": " << is_fixed
<< "\n";
}
// write the covariance matrix (omitting fixed parameters)
yaml_file << "covariance:\n";
for (unsigned int i = 0; i < cov.Nrow(); ++i) {
yaml_file << yaml_indent << mnState.Name(parNo[i]) << ":\n";
for (unsigned int j = 0; j < cov.Nrow(); ++j) {
yaml_file << yaml_indent << yaml_indent << mnState.Name(parNo[j])
<< ": " << cov(i, j) << "\n";
}
}
// write the correlation matrix (omitting fixed parameters)
yaml_file << "correlation:\n";
for (unsigned int i = 0; i < cov.Nrow(); ++i) {
yaml_file << yaml_indent << mnState.Name(parNo[i]) << ":\n";
for (unsigned int j = 0; j < cov.Nrow(); ++j) {
double correlation =
i == j ? 1.0
: cov(i, j) / (fMnUserParams.Error(parNo[i]) *
fMnUserParams.Error(parNo[j]));
yaml_file << yaml_indent << yaml_indent << mnState.Name(parNo[j])
<< ": " << correlation << "\n";
}
}
// close the yaml file
yaml_file.close();
} }
parNo.clear(); // clean up parNo.clear(); // clean up
} else { } else {