Alternative API?

[John Myles White]

I think it would be great to strive for a simple API that takes univariate and multivariate distributions defined by a density and returns samples. For example, a function called mcmc() that takes as inputs:

• log_density: The actual log density function, which takes a vector of parameters.
• initial_parameters: The initial value of the parameters for all chains. Recycled if only one vector is supplied.
• parameter_bounds: Specify hard lower and upper bounds on the parameters. Defaults to -Inf and +Inf everywhere.
• sampler_type: A Julia type specify the Metropolis sampler, the slice sampler, etc.
• desired_samples: The number of samples desired in total per chain.
• n_chains: The number of chains to run.
• thinning: The frequency with which samples are maintained from each sequence of draws.