Samplers

pypbl.samplers

Samplers for Bayesian inference.

pypbl.samplers.ensemble_sampler(fun, start, sigma, iterations, verbose=False)

Sampler based on the affine-invariant ensemble sampler for Markov chain Monte Carlo

Parameters:

• fun (function) – log probability function used to infer weights.
• start (list) – initial weights (it is recommended to use the MAP estimate).
• sigma (float) – term used to encourage different starting conditions for walkers.
• iterations (int) – number of iterations in sampling algorithm (total number of evaluation will be 2 * number of weights * iterations). # noqa
• verbose (boolean) – set as True to get verbose print out.

Returns:

Numpy array of samples

pypbl.samplers.simple_sampler(fun, start, sigma, iterations, verbose=False)

Simple sampler based on the metropolis hastings algorithm for Markov chain Monte Carlo

Parameters:

• fun (function) – log probability function used to infer weights
• start (list) – initial weights (it is recommended to use the MAP estimate)
• sigma (float) – diagonal term for proposal distribution covariance
• iterations (int) – number of iterations in sampling algorithm
• verbose (boolean) – set as True to get verbose print out

Returns:

Numpy array of samples