This is a sub-project of Peer-to-Peer AI Tracing App delegated by Prof. Yoshua Bengio. Read more about the app in Prof. Bengio's blog post.
The simulator is built using simpy.
It simulates human mobility along with infectious disease (COVID) spreading in a city, where city has houses, grocery stores, parks, workplaces, and other non-essential establishments.
Human mobility simulation is based on Spatial-EPR model. More details on this model are here and here.
The infection spread in this simulator is modeled according to what is known about COVID-19. The assumptions about the COVID-19 spread and mobility implemented in the simulator are in the Google Doc. The same document also details our current understanding of COVID-19. Our understanding is based on the published research as well as interactions with the epidemiologists. We plan to update the simulator as more and more about COVID-19 will be known.
Following python packages are required (python>=3.7)
pip install -r requirements.txt
Run the simulator as -
python run.py sim --n_people 100 --init_percent_sick 0.01 --seed 0
The simulator will output a logfile to output/sim_people-{N_PEOPLE}_days-{SIMULATION_DAYS}_init-{INIT_PERCENT_SICK}_seed-{SEED}_{DATE}-{TIME}/data.zip. It is a .zip file of list of dict pickles which contains a log of the mobility activity of a population of humans in simulator.py.
Run the risk prediction algorithms as -
python risk_prediction.py
This file reads in the logs that are output from the simulator, and runs a risk prediction algorithm based on:
- The reported symptoms of that individual, given they have the app.
- The encounters that individual had (which contain quantized user ids and risk levels).
- The risk update messages that are sent when a previously encountered user's risk level changes significantly.
You may select the model which you want to specify, or add your own model in that file. This file also handles clustering of messages (under the currently understood privacy model), which can be useful for risk prediction algorithms which depend on knowing how many contacts happened with one individual.
Run -
python run.py test
@click.option('--n_people', help='population of the city', type=int, default=100)
@click.option('--init_percent_sick', help='% of population initially sick', type=float, default=0.01)
@click.option('--simulation_days', help='number of days to run the simulation for', type=int, default=30)
@click.option('--out_chunk_size', help='number of events per dump in outfile', type=int, default=2500, required=False)
@click.option('--outdir', help='the directory to write data to', type=str, default="output", required=False)
@click.option('--seed', help='seed for the process', type=int, default=0)
Load the output of the simulator as following
data = pickle.load(open("output/data.pkl", 'rb'))
Although not designed with this usage in mind one can still call it like this
from run import run_simu
monitors = run_simu(n_stores=100, n_parks=50, n_people=100, n_misc=100, init_percent_sick=0.01, print_progress=False, seed=0)
data is a list of dict.
Following will require cufflinks and plotly.
python run.py base
It will open a browser window with the plot of SEIR curves.
Human class builds people, and Location class builds stores, parks, workplaces, households, and non-essential establishments.
data is a list. Each entry in the list is an event represented as a dict.
The detailed information about events is in docs/events.md
Please get in touch with me at pgupta@turing.ac.uk. There are several people working on it, so it will be the best use of everyone's time and effort if we all work on different aspects of this project.
Some areas that need work are listed here. We track and manage our tasks using Google Sheets.
@marco-gires, @marie-pellat, @teganmaharaj, @giancarlok, @thechange, @soundarya98, @mweiss17