como_command_line

Command-line version of the CoMo model.

Requirements

Tests were run on R version 4.1.0. For the CPP version of the solver (much faster), one needs to install the comoOdeCpp package, version 16.6.0. From within R, this can be installed like so:

R -e 'library(devtools); install_github("p-robot/comoOdeCpp@new_variant", subdir="comoOdeCpp")'

# The original CPP folder is found here (but does not include the ability to adjust transmission from introduction of a new variant): 
R -e 'library(devtools); install_github("bogaotory/comoOdeCpp@v16.6.0", subdir="comoOdeCpp")'

The CoMo model uses the following R packages: tidyverse, deSolve.

If the model is to be run on an AWS Lightsail, then the script lightsail_setup.sh includes all commands needed to prepare the lightsail node for deploying the command-line version of CoMo. It may be required to define a library for R packages:

# Make personal library for R packages if necessary
mkdir "~/R/x86_64-pc-linux-gnu-library/4.1"

Basic usage

The main script como_command_line.R can be called from a terminal like so which will run the CoMo model for a particular country, using the specified template, and saving output to the desired CSV file:

Rscript como_command_line.R <country_name> <template_path> <output_file>

where the input arguments are the following:

• country_name: country name of interest
• template_path: file path to the 'baseline' CoMo template
• output_file: file path to the output CSV file

Options

• The CoMo model is solved using the CPP solver comoOdeCpp by default. Should one wish to use the R version of the solver (i.e. via the deSolve package - a slower solver), then adjust the option USE_CPP <- FALSE in the script `como_command_line.R.

• This version of the CoMo model also includes the ability to include the introduction of a new variant at a particular point in time, with a multiplier on the variable p (the probability of transmission given contact). These parameters are included in a new worksheet in the template called Extra Param. See the example in tests/data/Template_CoMoCOVID19-App_v17_variant.xlsx.

Parameter grid search

Generate parameter sample

A grid search across parameter space can be performed. A grid of parameters first needs to be generated which can be performed using the following call:

Rscript como_random_sample.R <seed> <calibrated_parameters_csv> <n_samples> <output_csv>

where the input arguments are the following:

• seed: Integer used to see the random number generator in R
• calibrated_parameters_csv: CSV file of parameters to generate a random grid search across (see example in parameters/calibrated_parameters.csv.
• n_samples: Number of samples to draw from a random grid.
• output_csv: CSV file of sample from parameter space (number of columns is number of parameters and n_samples rows).

Simulate CoMo model across parameter sample

Evaluating the model for each parameter value, adjusted from a baseline template, can be run in the following manner, which will save a new file in the output_dir, entitled como_output_row<i>.csv, for each row i of the parameter_csv file:

Rscript como_parallel.R <country_name> <template_path> <parameter_csv> <output_dir>

where the input arguments are the following:

• country_name: country name of interest
• template_path: file path to the 'baseline' CoMo template
• parameter_csv: path to a CSV file of parameter values over which to evaluate the como model (adjusted from the baseline template).
• output_dir: directory of where to save CoMo output files (this directory should exist already)

On AWS Lightsail, running the model across a grid search can be set running using the following (which will run in the background even if the user logs off):

nohup Rscript como_parallel.R > log/como_parallel.log 2> log/como_parallel.err &

Sending output to file log/como_parallel.log and error to file log/como_parallel.err.

Tests

Templates for v17 of the CoMo model are taken from the CoMo github release v17.2.0 (or here). The file Template_CoMoCOVID-19App_v17.xlsx is available for download from that link. This was process with V17.2 of the Shiny App to generate the output file COVID19_App_Data_Template_CoMoCOVID-19App_v17.csv against which the command-line invocation is compared.

All tests can be run using the script run_all_tests.sh.