idontgetoutmuch / nonparametric-mh

NPMH and NPLiftedMH

We compare our Turing implementation of the NPMH and NPLiftedMH samplers with Turing's built-in SMC sampler. Note that even through we have added a seed, Turing’s SMC implementation is not deterministic.

Getting started

1. Download and install Julia by following the instructions at https://julialang.org/downloads/.
2. Run julia from the command line to start a Julia interactive session (also known as a read-eval-print loop or "REPL").
3. Run ] add Turing, Random, Distributions, DataFrames, CSV, PlotlyJS to install essential Julia packages for our implementation.

Generating Samples using the SMC, NPMH and NPLiftedMH Samplers

Note that Turing's SMC implementation is nondeterministic, so its results may vary somewhat.

1. Run ] activate NPMH on the REPL to activate the NPMH package.
2. Run include("infinite_gmm_npmh.jl") on the REPL to sample from the infinite Gaussian mixture model using the SMC and NPMH samplers and store them in the data folder.
3. Run ] activate NPLiftedMH on the REPL to activate the NPLiftedMH package.
4. Run include("infinite_gmm_npmhp.jl") on the REPL to sample from the infinite Gaussian mixture model using the NPLiftedMH sampler and store them in the data folder.

Visualising the Samples

1. Run include("visualise.jl") on the REPL to plot the histogram of the posterior and store it in the images folder.