Lots of different methods that aim to determine the marginal likelihood of a model by running some sort of Monte Carlo routine, for example #160 #639
Note these are separate to samplers (for example, nested sampling) which return estimates of the marginal likelihood alongside posterior samples.
Speaking with @martinjrobins and @MichaelClerx, we decided that it was probably best to implement all these methods as a separate class of methods, similar to, for example, those in MCMC.
The idea would be to pass a log_likelihood and log_prior to a MarginalLikelihoodSampling/Controller object, then select a method amongst those available.
Idea is to start with a few examples (after PINTS paper) then perhaps get students to take up the reigns...
Lots of different methods that aim to determine the marginal likelihood of a model by running some sort of Monte Carlo routine, for example #160 #639
Note these are separate to samplers (for example, nested sampling) which return estimates of the marginal likelihood alongside posterior samples.
Speaking with @martinjrobins and @MichaelClerx, we decided that it was probably best to implement all these methods as a separate class of methods, similar to, for example, those in MCMC.
The idea would be to pass a log_likelihood and log_prior to a MarginalLikelihoodSampling/Controller object, then select a method amongst those available.
Idea is to start with a few examples (after PINTS paper) then perhaps get students to take up the reigns...