Multiple model runs¶
In the last page you successfully performed a single run modelling an outbreak of the lurgy that started in London. This run (which we call a model run) is stochastic, meaning that the results will be slightly different every time it is performed.
To gain confidence in any predictions, we need to perform a model run multiple times, and average over the results.
Performing multiple model runs¶
metawards has the command line option --repeats (or -r) to
set the number of times a model run should be repeated. For example,
run the below command to repeat the model run four times;
metawards -d lurgy -a ExtraSeedsLondon.dat --repeats 4
metawards will automatically use as many of the cores on your computer
as it can to parallelise the jobs. On my computer, the output shows;
Performing 4 runs of each set of parameters
Number of threads to use for each model run is 1
Number of processes used to parallelise model runs is 4
Parallelisation will be achieved using multiprocessing
I have four processor cores on my laptop, so I see the four repeats run
in parallel using four processes, with each model run performed
using 1 thread. You will see a different distribution of threads
and processes if you have a different number of cores on your computer.
You can set the number of processes that metawards should use via
the --nprocs command line option. You can set the number of threads
that metawards should use via the --nthreads command line option.
This calculation may take some time (2-3 minutes). This time, instead
of seeing a summary of the outbreak, metawards will show a summary
of the different model run jobs. Something similar to this should
be printed;
Running 4 jobs using 4 process(es)
Running jobs in parallel using a multiprocessing pool...
Completed job 1 of 4
(NO_CHANGE)[repeat 1]
2020-12-19: DAY: 243 S: 11776504 E: 0 I: 0 R: 44305573 IW: 1 TOTAL POPULATION 56082077
Completed job 2 of 4
(NO_CHANGE)[repeat 2]
2020-12-16: DAY: 240 S: 11787147 E: 0 I: 0 R: 44294930 IW: 0 TOTAL POPULATION 56082077
Completed job 3 of 4
(NO_CHANGE)[repeat 3]
2020-11-25: DAY: 219 S: 11789948 E: 0 I: 0 R: 44292129 IW: 0 TOTAL POPULATION 56082077
Completed job 4 of 4
(NO_CHANGE)[repeat 4]
2020-12-04: DAY: 228 S: 11782418 E: 0 I: 0 R: 44299659 IW: 1 TOTAL POPULATION 56082077
Writing a summary of all results into the
csv file /Users/chris/GitHub/tutorial/output/results.csv.bz2.
In this case, all four outbreaks completed within 219-243 days, while the number of the population who progressed to the ‘R’ state were all around 44.3 million.
The results.csv.bz2 file¶
The day-by-day progress of each the outbreak for each model run is
recorded in the output file results.csv.bz2. This is a comma-separated
file that has been compressed using bzip2.
You can read this file easily using
Python Pandas or with
R. You can even import this into Excel
(although you may need to uncompress this file first using bunzip2).
For example, if you have Pandas installed, then you can read this file via an ipython or Jupyter notebook session via;
>>> import pandas as pd
>>> df = pd.read_csv("output/results.csv.bz2")
>>> df
fingerprint repeat day date S E I R IW
0 NO_CHANGE 1 0 2020-04-20 56082077 0 0 0 0
1 NO_CHANGE 1 1 2020-04-21 56082077 0 0 0 0
2 NO_CHANGE 1 2 2020-04-22 56082072 5 0 0 0
3 NO_CHANGE 1 3 2020-04-23 56082072 0 5 0 0
4 NO_CHANGE 1 4 2020-04-24 56082068 0 5 4 4
.. ... ... ... ... ... .. .. ... ..
929 NO_CHANGE 4 224 2020-11-30 11782419 0 4 44299654 0
930 NO_CHANGE 4 225 2020-12-01 11782419 0 3 44299655 0
931 NO_CHANGE 4 226 2020-12-02 11782419 0 1 44299657 0
932 NO_CHANGE 4 227 2020-12-03 11782419 0 1 44299657 0
933 NO_CHANGE 4 228 2020-12-04 11782418 0 0 44299659 1
[934 rows x 9 columns]
Each repeat is given its own number, which is in the repeat column.
The day of the outbreak is given in the day column. This counts up
from day zero when the outbreak started, to the last day when the
outbreak was over. You can control the start day of the outbreak using
the --start-day command line option.
The date column contains the date of each day in the outbreak. By
default, metawards assumes that day zero is today. You can set the
date of day zero using the --start-date command line option, e.g.
--start-date tomorrow would start tomorrow, while
--start-date Jan 1 would start on January 1st this year.
The values of S, E, I, R and IW for each repeat for each day are then given in their correspondingly named columns.
The fingerprint column not used for this calculation - we will see what it is later.