chrism0dwk / covid19uk Goto Github PK

@chrism0dwk, I was thinking of implementing some convergence diagnostics for your MCMC; is this useful? I'm thinking effective sample size and maybe Gelman-Rubin R-hat.

https://arxiv.org/pdf/1909.11827.pdf

In order for the model to be used in the model runner, it needs to be containerized. Brief instructions are here. Here's an example for IDMs covasim:

FROM python:3

ARG COVASIM_VER=1.0.0
RUN pip install covasim==${COVASIM_VER} pandas

COPY . /app

CMD ["python", "/app/runsim.py", "/data/input/inputFile.json", "/data/output/data.json"]

Models in the model runner take a JSON input and generate JSON output, which is validated using a JSON schema. The Typescript classes are a little easier to read:

Input
Output

It should be straightforward to take the minimal input and use it to modify a YAML template that you currently use. It doesn't look as though you use intervention data?

I am using the reticulate package that allows you call Python from R. This is how the first part of covid_ode.py looks like in R:

library(reticulate)
covid <- import("covid")
config <- yaml::read_yaml("ode_config.yaml")

# Load age mixing 
K_tt <- readRDS(config$data$age_mixing_matrix_term)

# Load age mixing holiday
K_hh <- readRDS(config$data$age_mixing_matrix_hol)

# Load mobility matrix 
Tmat <- readRDS(config$data$mobility_matrix)
diag(Tmat) <- 0 

N <- readRDS(config$data$population_size)
 
param <- config$parameter
settings <- covid$util$sanitise_settings(config$settings)


model <- covid$model$CovidUKODE(K_tt, K_hh, Tmat, N, 
                                settings$start, settings$end, settings$holiday,
                                settings$bg_max_time, 1)

Note that it won't run because there are no start and end settings in the ode_config.yaml.

A function like mcmc.py can be run from R as simply as:

source_python("mcmc.py")

It would be useful to have a chat to know what arguments would you like to have in a R wrapper for each of the main functions.

Hi @chrism0dwk

This is a high-level issue to track what is needed in order to integrate the model with https://github.com/covid-policy-modelling/model-runner. I'll post separate issues for each of these.

Accept input and generate output
Create Dockerfile

Function load_commute_volume in pydata.py doesn't work correctly.
The sort_index sorts by day not date so it ends up in the wrong order and commute.merge doesn't correctly match several dates.
Adding "commute_raw.index = pd.to_datetime(commute_raw.index, format = '%d/%m/%Y')" just after "commute_raw = pd.read_csv(filename, index_col='date')", changes the index to a datetime format, allowing the sorting and merging to work correctly.