Day 3

Note

Set up environment, corrected file paths, add timer, understood loops, and rain main.R as provided. Total time used: 2h 1m (5.0%)

12.13-12.26, 12.30-12.31: Set up environment and copy over files

I created an .RProj in reproduction/, then looked to the repository and article for any packages and versions.

• Supplementary material: R 3.6.0
• Article: No mention
• Repository: master.R has library(tidyverse) and covid_triage_simr.R has:

library(parallel)
library(tidyr)
library(dplyr)
library(ggplot2)
library(gridExtra)

As I have had difficulty attempting to backdate R, I will first try this in my current version of R (4.4.1) with the latest packages.

I created a DESCRIPTION file with these packages:

Title: Computational reproducibility assessment of Wood et al. 2021
Depends: 
    R
Imports:
    parallel,
    tidyr,
    dplyr,
    ggplot2,
    gridExtra,
    tidyverse

I set up the renv:

renv::init(bare=TRUE)
renv::install()
renv::snapshot()

I copied the scripts and inputs into reproduction/, but not the original outputs. I reorganised these into inputs/ and scripts/ folders.

I looked over the scripts, easily identifying master.R (which I renamed to main) as the primary script to run, which itself uses covid_triage_simr.R which contains the model code. There were several scripts beginning fig... for creating plots from the article, it appears.

Then spotted cowplot dependency in fig .R scripts, which add to renv.

13.31-13.36, 13.55-14.18, 14.34-14.46: Running main.R

Before running, amended the file paths (with intention of running file from the reproduction/ folder) and commented out setwd() (which had been provided blank (xxx) for users to fill in with their path).

I ran main.R. The article stated it should take 5 minutes per 1000 replications. I found it seemed to take a little more like 10. The output print statements appeared like this:

`summarise()` has grouped output by 'metric'. You can override using the `.groups` argument.
`summarise()` has grouped output by 'dates', 'group'. You can override using the `.groups` argument.
`summarise()` has grouped output by 'dates', 'group'. You can override using the `.groups` argument.
Joining with `by = join_by(dates, group, metric, currency)`
[1] completed: cyclical_1_NA_TRUE_TRUE_TRUE_TRUE_TRUE_TRUE
`summarise()` has grouped output by 'metric'. You can override using the `.groups` argument.
`summarise()` has grouped output by 'dates', 'group'. You can override using the `.groups` argument.
`summarise()` has grouped output by 'dates', 'group'. You can override using the `.groups` argument.
Joining with `by = join_by(dates, group, metric, currency)`
[1] completed: cyclical_1_NA_TRUE_TRUE_TRUE_TRUE_TRUE_FALSE
`summarise()` has grouped output by 'metric'. You can override using the `.groups` argument.
`summarise()` has grouped output by 'dates', 'group'. You can override using the `.groups` argument.
`summarise()` has grouped output by 'dates', 'group'. You can override using the `.groups` argument.
Joining with `by = join_by(dates, group, metric, currency)`
[1] completed: cyclical_1_NA_TRUE_TRUE_TRUE_TRUE_FALSE_FALSE

I was worried it might actually take a long time to run, so looked into what it was looping through (i.e. how many).

for (cap in seq(10,200,10)) { - loops through 20 capacities from 10 to 200, which are fed into covid_simr2.

for (scenario in unique(cases_raw$scenario)) { - where unique(cases_raw$scenario) is "cyclical" "lockdown" "unmitigated".

for (policy in 1:3) { - loops through 1 2 3, which sets:

• Policy 1: policy_param_ls<-NA
• Policy 2: policy_param_ls<-seq(3,6,3)
• Policy 3: policy_param_ls<-NA

for (policy_param in policy_param_ls) { - so either looping through NA (and so only running once) or seq(3,6,3) which is 3 6 (with policy_param being an input to covid_simr2.). Within this, it sets crit_ls1<-crit_ls is policy is 1, or crit_ls1<-crit_ls[-1] if it is otherwise. crit_ls is a list defined at the start with a series of true/false:

crit_ls[[1]]<-c(T,T,T,T,T,T)
crit_ls[[2]]<-c(T,T,T,T,T,F)
crit_ls[[3]]<-c(T,T,T,T,F,F)
crit_ls[[4]]<-c(T,T,T,F,F,F)
crit_ls[[5]]<-c(T,T,F,F,F,F)

for (crit in crit_ls1) { then loops through crit_ls1 and runs the model.

Hence, in the 20/25 minutes or so that I got :

[1] completed: cyclical_1_NA_TRUE_TRUE_TRUE_TRUE_TRUE_TRUE
[1] completed: cyclical_1_NA_TRUE_TRUE_TRUE_TRUE_TRUE_FALSE
[1] completed: cyclical_1_NA_TRUE_TRUE_TRUE_TRUE_FALSE_FALSE

I had actually only completed a very small number of scenarios. I wrote those three results to csv using the final lines of the loop. The outp_raw result is very large.

I decided to simply set this to run on the remote machine. Before doing so, I add some code to record and save time elapsed.

I cloned it onto the machine, ran:

Rscript -e "renv::restore()"

And then:

Rscript -e "renv::run('scripts/main.R')"

Timings

import sys
sys.path.append('../')
from timings import calculate_times

# Minutes used prior to today
used_to_date = 67

# Times from today
times = [
    ('12.13', '12.26'),
    ('12.30', '12.31'),
    ('13.31', '13.36'),
    ('13.55', '14.18'),
    ('14.34', '14.46')]

calculate_times(used_to_date, times)

Time spent today: 54m, or 0h 54m
Total used to date: 121m, or 2h 1m
Time remaining: 2279m, or 37h 59m
Used 5.0% of 40 hours max