class RooStats::ProfileLikelihoodCalculator: public RooStats::CombinedCalculator
ProfileLikelihoodCalculator is a concrete implementation of CombinedCalculator (the interface class for a tools which can produce both RooStats HypoTestResults and ConfIntervals). The tool uses the profile likelihood ratio as a test statistic, and assumes that Wilks' theorem is valid. Wilks' theorem states that -2* log (profile likelihood ratio) is asymptotically distributed as a chi^2 distribution with N-dof, where N is the number of degrees of freedom. Thus, p-values can be constructed and the profile likelihood ratio can be used to construct a LikelihoodInterval. (In the future, this class could be extended to use toy Monte Carlo to calibrate the distribution of the test statistic).
Usage: It uses the interface of the CombinedCalculator, so that it can be configured by specifying:
- a model common model (eg. a family of specific models which includes both the null and alternate),
- a data set,
- a set of parameters of interest. The nuisance parameters will be all other parameters of the model
- a set of parameters of which specify the null hypothesis (including values and const/non-const status)
After configuring the calculator, one only needs to ask GetHypoTest() (which will return a HypoTestResult pointer) or GetInterval() (which will return an ConfInterval pointer).
The concrete implementations of this interface should deal with the details of how the nuisance parameters are dealt with (eg. integration vs. profiling) and which test-statistic is used (perhaps this should be added to the interface).
The motivation for this interface is that we hope to be able to specify the problem in a common way for several concrete calculators.
Function Members (Methods)
| void | DoGlobalFit() const |
| void | DoReset() const |
| RooAbsData* | RooStats::CombinedCalculator::GetData() const |
| RooAbsPdf* | RooStats::CombinedCalculator::GetPdf() const |
Data Members
| RooFitResult* | fFitResult | internal result of gloabl fit |
| RooArgSet | RooStats::CombinedCalculator::fAlternateParams | RooArgSet specifying alternate parameters for hypothesis test // Is it used ???? |
| RooAbsData* | RooStats::CombinedCalculator::fData | |
| RooArgSet | RooStats::CombinedCalculator::fNuisParams | RooArgSet specifying nuisance parameters for interval |
| RooArgSet | RooStats::CombinedCalculator::fNullParams | RooArgSet specifying null parameters for hypothesis test |
| RooArgSet | RooStats::CombinedCalculator::fPOI | RooArgSet specifying parameters of interest for interval |
| RooAbsPdf* | RooStats::CombinedCalculator::fPdf | |
| Double_t | RooStats::CombinedCalculator::fSize | size of the test (eg. specified rate of Type I error) |
Class Charts
Function documentation
constructor from pdf and parameters the pdf must contain eventually the nuisance parameters
construct from a ModelConfig. Assume data model.GetPdf() will provide full description of model including constraint term on the nuisances parameters
destructor
cannot delete prod pdf because it will delete all the composing pdf's
if (fOwnPdf) delete fPdf;
fPdf = 0;
reset and clear fit result to be called when a new model or data are set in the calculator
perform a global fit of the likelihood letting with all parameter of interest and nuisance parameters keep the list of fitted parameters
Main interface to get a RooStats::ConfInterval. It constructs a profile likelihood ratio and uses that to construct a RooStats::LikelihoodInterval.
Main interface to get a HypoTestResult. It does two fits: the first lets the null parameters float, so it's a maximum likelihood estimate the second is to the null (fixing null parameters to their specified values): eg. a conditional maximum likelihood the ratio of the likelihood at the conditional MLE to the MLE is the profile likelihood ratio. Wilks' theorem is used to get p-values