Parsing EpiFusion Outputs

What is in the output?

For each MCMC chain, EpiFusion will give you the following output files:

• betas: csv file where each row is a trajectory of rate beta sampled from the MCMC
• trajectories: csv file where each row is a daily infection trajectory sampled from the MCMC
• rt: csv file where each row is a daily Rt trajectory sampled from the MCMC
• params: txt file where each column is an MCMC parameter, and each row is an MCMC sample
• likelihoods: txt file of the posterior likelihoods from each MCMC step
• acceptance: txt file where each line logs the acceptance rate of steps between MCMC samples
• completed: txt file where each line logs if the particle filter step was completed or quit due to particle depletion
• positivetests (only for combined or epi-only analyses): csv file where each row is simulated case incidence by the model which was compared to the observed case incidence

EpiFusion will also save a copy of the parameter file used to the output folder, so you can remember exactly what parameters were used, and a file called 'timings.txt' with the runtime in nanoseconds.

How do I interpret the output?

We have an R package on GitHub called 'EpiFusionUtilities' which can be useful for interpreting the output. It can be installed via devtools with the following command in R:

devtools::install_github("https://github.com/ciarajudge/EpiFusionUtilities")

For more on EpiFusionUtilities, visit the project wiki.

Main EpiFusionUtilities Functions for Parsing Outputs

load_raw_epifusion("path_to_outputfolder")

• Description: Loads an R list with the full posterior trajectory samples, parameter samples, likelihoods and acceptance rates for all chains.
• Input: A filepath to an EpiFusion outputs folder
• Output: A large list with the full EpiFusion output in an R compatible format.

raw_epifusion_output <- load_raw_epifusion('epifusionoutputs/')

plot_likelihood_trace(raw_epifusion_output)

• Description: Plots the likelihood trace for all chains in an EpiFusion analysis
• Input: Raw epifusion output object generated by load_raw_epifusion()
• Output: A plot of the likelihood trace coloured by chain.

plot_likelihood_trace(raw_epifusion_output)

extract_posterior_epifusion(raw_epifusion_output, burn_in)

• Description: Extracts the final epifusion posterior samples, discarding a certain proportion at the beginning of each chain as burn-in
• Input: Raw epifusion output object generated by load_raw_epifusion() and proportion of each chain to discard as burn-in
• Output: A large list with the EpiFusion posterior samples aggregated across chains (with burn-in discarded). Also has trajectory means and HPDs, and parameter convergence statistics.

extract_posterior_epifusion(raw_epifusion_output, 0.1)