MetropolisSampler

class agabpylib.mcmc.metropolis.MetropolisSampler(lnprob, proposal_rvs)

Bases: object

Base class for Metropolis samplers. Can deal with multi-dimensional sampling problems.

Limitations:

• Simplistic implementation

• Only one chain at a time

• No conservation of state

Methods Summary

get_acceptance_fraction()	Get the acceptance fraction for the MCMC chain.
get_samples()	Get the MCMC samples.
run_mcmc(theta_init, n_iter[, burnin, thin])	Run the sampler.

Methods Documentation

get_acceptance_fraction()

Get the acceptance fraction for the MCMC chain.

Parameters:

None –

Returns:

acceptance_frac – The acceptance fraction.

Return type:

float

get_samples()

Get the MCMC samples.

Parameters:

None –

Returns:

samples – The array with MCMC samples (shape (n_samples, n_parameters)).

Return type:

float array

run_mcmc(theta_init, n_iter, burnin=0, thin=1)

Run the sampler.

TODO: Handle the case where the probability densities for both the current and proposed sample are -np.inf.

Parameters:

• theta_init (float, or float array) – Initial values of the samples theta_k.

• n_iter (int) – Number of MCMC iterations to perform.

• burnin (int) – Number of “burn-in” steps to take.

• thin (int) – Only output every thin samples.

Return type:

Nothing