This repository contains the nextflow scripts accompanying the paper "Parallel Tempering With a Variational Reference" by Surjanovic, Syed, Bouchard-Côté, and Campbell. NeurIPS (2022). https://arxiv.org/abs/2206.00080

See https://github.com/UBC-Stat-ML/bl-vpt for the source code of the software benchmarked by the scripts in the present repository.

Required software

• bash
• the following should be available in the PATH variable:
  • Java Open SDK, version 11 (use sdkman)
  • Rscript
• the following R packages should be installed:
  • ggplot2
  • dplyr
  • ggridges

Replicating the optimization (section 4.2) and PT topology (4.4) results

Results will be written in a subdirectory of ptgrad-nextflow/deliverables.

For the PT topology comparison, use ./nextflow run pt_topologies.nf -resume

Due to the various factorial designs, grid searches and replicates, these experiments are best performed on a cluster. For example, if your cluster uses a PBSPRO scheduler and supports singularity and conda, on the login node, use

./nextflow-pbspro.sh run optimization.nf -resume

which will submit for you to the queue parallel runs via qsub, gather the result and create the plots.

For other types of cluster schedulers, edit pbspro.config based on the nextflow documentation.