### Data for "Efficient Bayesian inference of fully stochastic epidemiological models with applications to COVID-19."
#### Yuting I. Li et al., [Roy. Soc. Open Sci. 8, 211065 (2021)](https://doi.org/10.1098/rsos.211065).
see also the preprint [arXiv:2010.11783](https://arxiv.org/abs/2010.11783) and [github data repo](https://github.com/rljack2002/infExampleCovidEW)
### Overview
This repository contains python code (scripts and jupyter notebooks) for generation and processing of the results of the paper. Results are obtained using the [PyRoss](https://github.com/rajeshrinet/pyross) library to analyse an epidemiological compartment model for COVID-19 in England and Wales, during March-May 2020.
Running the code requires `PyRoss`, see https://github.com/rajeshrinet/pyross
### Repo contents
The folder `SimpleTestModel` contains a simple model to validate the inference methodology (see the README.md file in that directory)
Other folders are elevant for the model for COVID-19 in England/Wales (the names indicate the sources used for *contact matrices*)
#### Figures
are in the folder `finalFigs/`, they are generated by python scripts as detailed below
#### other folders (subdirectories)
are relevant for model variants
* `fumanelli-step` : $`C^F`$ variant, step-like NPI
* `prem-step` : $`C^P`$ variant
* `mix-step` : $`C^M`$ variant
* `fumanelli-ez` : $`C^F`$ variant, NPI-with-easing
#### Notebooks
In most cases, the notebook "code" is identical in each folder, the results are different because variant
information is loaded from `expt_params_local.py`. The only small differences are for the MCMC analysis of the `-ez-` variant,
when displaying MCMC traces.
To understand PyRoss syntax, see the [ documentation](https://pyross.readthedocs.io/en/latest/) and [examples](https://github.com/rajeshrinet/pyross#examples)
Notebooks that exist in every folder.
* ew-inf.ipynb : parameter optimisation for MAP
* ew-mcmc.ipynb : MCMC on posterior
* ew-mcmcPost.ipynb : MCMC post-process
* ew-evidence.ipynb : evidence computation
Notebooks only for `fumanelli-step` and `fumanelli-ez`
* ew-sample.ipynb : stochastic forecasts
Notebooks only for `fumanelli-step`
* ew-FIM.ipynb : Fisher information matrix
Additional notebooks for `fumanelli-ez`
* ew-tWinNN-inf.ipynb : inference : where NN=09,10,11 is (inference window length + 1)
* ew-tWinNN-mcmc.ipynb : ditto for mcmc
* ew-tWinNN-result_mcmc.ipynb : ditto for mcmc postprocess
#### Data
Data that exist in each folder
* ewMod-inf.pik : inference result
* ewMod-result_mcmc.pik : representative MCMC samples
(note: the full set of MCMC samples (`ewMod-mcmc.pik`) is *not* provided here due to very large file sizes, but `ewMod-result_mcmc.pik` contains representative samples)
Data only for `fumanelli-step` and `fumanelli-ez`
* ewMod-sample-post.pik : stochastic forecast (posterior sampled params)
* ewMod-sample-map.pik : stochastic forecast (MAP params)
Data only for `fumanelli-step`
* ewMod-FIM.npy : Fisher information matrix (as numpy array)
* ewMod-FIM-evec.pik : spectrum of FIM
Figures are generated by python scripts
(these also use PyRoss to run deterministic forecasts, etc, they should be run direct from the directories where they live)
* `fumanelli-step/stepFigs.py` : graphs for step-like NPIs and for model comparison
* `fumanelli-ez/easeFigs.py` : graphs for NPI-with-easing
* `fumanelli-ez/genFigs.py` : general figs, NPIs, contact matrices, etc