sampler_nuts: No U-Turn Sampler.

In lucas-castillo/SampleR: Tools To Compare Human Performance To Sampling Algorithms

Description

Adapted from Hoffman and Gelman (2014). The No U-Turn Sampler (NUTS) aims to eliminate the need to set a number of steps L that is present in Hamiltonian Monte Carlo, which may lead to undesirable behaviour in HMC if not set correctly.NUTS does so by recursively building a set of candidate points that span the target distribution, and stopping when it starts to double back (hence its name). More information can be found here

Usage

sampler_nuts(pdf, start, epsilon, iterations = 1024, delta_max = 1000)

Arguments

pdf	Probability Density Function of the target distribution.Takes 2 arguments: a vector with a position, and, optionally, a boolean determining whether to return the log density instead.
start	starting point
epsilon	step size. If left to the default value, a suitable step-size will be estimated
iterations	Times the sampler runs
delta_max	Measure of the required accuracy of the simulation. The authors recommend a large value (1000)
iterations_adapt	burn-in period
delta	desired acceptance rate (default at 0.6)

Value

matrix of size iterations x dimensions with the points visited by the sampler.

Examples

pd_func <- make_distr_pdf(distr::Norm())
nuts_results <- sampler_nuts(pd_func, start = 0, iterations = 20)

lucas-castillo/SampleR documentation built on Jan. 1, 2021, 8:25 a.m.