sms's Introduction

smssample

This repo contains the scripts used in the following paper:

Distributed Bayesian Posterior Sampling via Moment Sharing

Minjie Xu, Balaji Lakshminarayanan, Yee Whye Teh, Jun Zhu and Bo Zhang

Advances in Neural Information Processing Systems (NIPS), 2014.

Please cite the above paper if you use this code.

If you have any questions/comments/suggestions, please contact Minjie ([email protected]) or Balaji ([email protected]).

Code released under MIT license (see LICENSE for more info).

Note: spikeslab_gibbs.m has been modified from Michalis Titsias's paired Gibbs sampler for Bayesian sparse linear regression with spike and slab prior.

List of scripts in the folder:

smssample.m (main script that does the bulk of the computation)

Examples of approximating families:

approxfam_gaussian.m
approxfam_spikeslab.m

Examples of model specification:

gaussian.m, gaussian_gendata.m
bayeslogreg.m, bayeslogreg_gendata.m
spikeslab.m, spikeslab_gendata.m
sparsebayeslogreg.m, sparsebayeslogreg_gendata.m

Examples of base samplers:

sampler_nuts_da.m
sampler_spikeslab.m (calls spikeslab_gibbs.m)

Utilities:

nuts_da.m
nuts_da.cpp
sigmoid.m
vislsm.m
viserrs.m

Demo scripts:

bayeslogreg_test.m (Bayesian logistic regression)
spikeslab_test.m (Bayesian linear regression with spike-and-slab prior over weights)

Note: You may need to write additional scripts to aggregate the results and generate the plots.

How do I use your scripts for my model/sampler?

See above for examples of model specifications, approximating families and samplers.

Create your model specification MODEL.m
Create your approximating family approxfam_XYZ.m
Create your sampler sampler_ABC.m
Create a wrapper script that will invoke the above and call smsample.m (see demo scripts above)

Recommend Projects