class RooStats::MetropolisHastings: public TObject

This class uses the Metropolis-Hastings algorithm to construct a Markov Chain of data points using Monte Carlo. In the main algorithm, new points in the parameter space are proposed and then visited based on their relative likelihoods. This class can use any implementation of the ProposalFunction, including non-symmetric proposal functions, to propose parameter points and still maintain detailed balance when constructing the chain.

The "Likelihood" function that is sampled when deciding what steps to take in the chain has been given a very generic implementation. The user can create any RooAbsReal based on the parameters and pass it to a MetropolisHastings object with the method SetFunction(RooAbsReal&). Be sure to tell MetropolisHastings whether your RooAbsReal is on a (+/-) regular or log scale, so that it knows what logic to use when sampling your RooAbsReal. For example, a common use is to sample from a -log(Likelihood) distribution (NLL), for which the appropriate configuration calls are SetType(MetropolisHastings::kLog); SetSign(MetropolisHastings::kNegative); If you're using a traditional likelihood function: SetType(MetropolisHastings::kRegular); SetSign(MetropolisHastings::kPositive); You must set these type and sign flags or MetropolisHastings will not construct a MarkovChain.

Also note that in ConstructChain(), the values of the variables are randomized uniformly over their intervals before construction of the MarkovChain begins.