Evaluating the impact of modelling strain dynamics on short-term COVID-19 forecast performance

This repository contains the code, and analysis used to evaluate the impact of modelling strain dynamics on short-term COVID-19 forecast performance. In addition it also contains the write up of this work.

Resources

• writeups: Summary paper and additional supplementary information.
• _targets.R and targets: Analysis workflow and modular sections.
• data: Input data and summarised output generated by steps in the analysis.
• Rendered summary paper.
• Rendered SI.
• analyses: Ad-hoc analyses not part of the overarching worflow.
• NEWS.md: Dated development notes.
• .devcontainer: Resources for reproducibility using vscode and docker.

Reproducibility

Dependencies

This project uses renv to store its package dependencies. This means that any call to R will result in the project trying to bootsrap its dependencies. Alternatively this project comes with a .devcontainer for use with vscode and or GitHub codespaces. Finally, if desired the supplied Dockerfile may be used to recreate the development environment.

Optionally a list of dependencies may be generated using the following:

renv::dependencies() 

If cmdstanr has not been installed previously then stan may need to be installed. This can be done using the following,

cmdstanr::install_cmdstan()

Analyses

All analyses have been implemented using a targets workflow in _targets.R with modules stored in targets. The full analysis can be recreated using the following,

. _targets.sh

Note that this is a computationally heavy pipeline and running it from end to end will require sufficient compute resources and some time. An archived version of this pipeline can be downloaded using the following,

Rscript get-targets-archive.R

The worflow can be explored using the following targets commands in an interactive R session.

• Run the workflow using all available workers.

targets::tar_make_future(workers = future::availableCores())

• Explore a graph of the workflow.

targets::tar_visnetwork(targets_only = TRUE)

• Watch the workflow as it runs.

targets::tar_watch(targets_only = TRUE)

To understand the workflow in more detail see the supplementary information and the summary paper.