Day 4 – Reproducing Johnson et al. 2021

Note

Attempted to run model with fewer agents but results 0/1/NaN so paused until have access to higher powered machine in a few weeks. Total time used: 3h 13m (8.0%)

09.27-09.52, 10.20-10.21, 10.40-10.43: Run model

Helpfully, we know to expect that this has a long run time from their README:

“Note: The analysis version of these results simulates 100 million agents, which will take ~16 hours to run for each script. The published analysis was conducted using R Version 3.5.3”

Given this warning, I tried running it with fewer agents:

settings$n_base_agents <- 5e+3

On my machine, I tried it with increasing numbers of agents, but it consistently returned results of 0 / 1 / NaN:

5000 (a minute to run)
50,000 (2.4 minutes to run)
500,000 (18.2 minutes to run)

Unfortunately I can’t currently access our remote higher spec machine, which would’ve allowed me to run these quicker.

I created an .Rmd file with only the first scenario, but set to 100 million agents as in the original analysis.

Timings

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

# Minutes used prior to today
used_to_date = 164

# Times from today
times = [
    ('09.27', '09.52'),
    ('10.20', '10.21'),
    ('10.40', '10.43')]

calculate_times(used_to_date, times)

Time spent today: 29m, or 0h 29m
Total used to date: 193m, or 3h 13m
Time remaining: 2207m, or 36h 47m
Used 8.0% of 40 hours max