tags: SCRC project-diary

SCRC Project Diary - 1

2020.07.02

This is the second iteration of my coding journey with the SCRC project. The last time, I looked into the infectionMap.txt alone which look like these:

13(0)         ->  [15(5)]
              ->  15(5)          ->  [14(8), 9100(14), 4217(16), 8133(18)]
                         ->  9100(14)       ->  [3993(24)]

and produced these graphics:
Graph with recurring random infections during the whole simulation [high-res pdf] Graph with no further random infections once the simulation started. The graphs are now much richer and longer [high-res pdf] An example of the longer chains 
Note: On Gephi, I follow the layout operations as follows:
- OpenOrd first which deals with the disconnected components nicely
- Fruchterman Reingold or Yifan Hu for laying them out in a spread
- And then Noverlap to resolve node overlaps

For the large graph, this is the distribution of components — there are 746 components:

Graph with no further random infections once the simulation started. The graphs are now much richer and longer
Graph with no further random infections once the simulation started. The graphs are now much richer and longer [high-res pdf]  

We also got a new Infection Map today with data containing locations. I gave that a try as well, there are 2108 components:
   

And here are the three largest 3 and 10 components from this graph:

Largest 3 components Largest 10 components

Additional Data — Inputs, Intermediate logs, etc.

Exploring the new datasets now that we got recently. These are pretty rich but one can only insert so much data on a graph. One idea is to go into connected multiple views where details are revealed in facets but that’s still an engineering question.

Events

events.csv which looks like this:

time","eventType","id","newStatus","additionalInfo"
0,"InfectionEvent",2048,"EXPOSED","This case was an initial infection"
0,"InfectionEvent",4101,"EXPOSED","This case was an initial infection"

where there are three different types of events : InfectionEvent, VirusEvent, AlertEvent

and the status changes follow the compartment structure that the underlying model follows. Here are the states, I guess these are what an epidemiologist will call compartments:

array(['EXPOSED', 'ASYMPTOMATIC', 'PRESYMPTOMATIC', 'RECOVERED',
       'SYMPTOMATIC', 'REQUESTED_TEST', 'SEVERELY_SYMPTOMATIC',
       'AWAITING_RESULT', 'ALERTED', 'TESTED_POSITIVE', 'TESTED_NEGATIVE',
       'NONE', 'DEAD'], dtype=object)

Note: Correction — some of these are compartments but some of them are about events, such as testing, etc.

Frequency of different infection events Frequency of different individual events

Timeline Stories There is an interesting timeline/story element to the data in this file. For instance, I’m tracing the “story” of node 15 which got infected by node 13 (first line in the infection map file):

6,"InfectionEvent",15,"EXPOSED","This case was due to contact with 13 at time = 5"
7,"VirusEvent",15,"PRESYMPTOMATIC","Old Status : EXPOSED"
11,"VirusEvent",15,"SYMPTOMATIC","Old Status : PRESYMPTOMATIC"
12,"AlertEvent",15,"REQUESTED_TEST",""
13,"VirusEvent",15,"SEVERELY_SYMPTOMATIC","Old Status : SYMPTOMATIC"
14,"AlertEvent",15,"AWAITING_RESULT",""
14,"AlertEvent",15,"TESTED_POSITIVE",""
18,"VirusEvent",15,"DEAD","Old Status : SEVERELY_SYMPTOMATIC"

Things didn’t end well for 15..

A question -- this seems like an error or a glitch:
There are nodes which are exposed twice
9,"InfectionEvent",2,"EXPOSED","This case was due to contact with 18 at time = 8"
9,"InfectionEvent",2,"EXPOSED","This case was due to contact with 2815 at time = 8"

The question here is whether there is an opportunity to reveal these timelines here? Does the timeline of a single individual matter? Probably not. So my guess is that an aggregation here would work pretty well. So I would like to think about a design on:

An aggregation on how the agents are changing states and their aggregate journeys through the states.

One idea could be to bin these events in terms of the transition types. The model description on the github page tells me that:

It comprises of S-E1-E2-Iasymp-Isymp-Isev-D-R disease progression compartments

S – Susceptible – the default condition E1 – Exposed, but not infectious E2 – Exposed, infectious, but not detectable by a test Ia – Infected and asymptomatic, will test positive Is – Infected and symptomatic, will test positive R – Recovered D – Dead

Which means there are transitions such as:

EXPOSED - PRESYMPTOMATIC
EXPOSED - ASYMPTOMATIC
PRESYMPTOMATIC - SYMPTOMATIC
SYMPTOMATIC - SEVERELY_SYMPTOMATIC
SYMPTOMATIC - RECOVERED
SEVERELY_SYMPTOMATIC - DEAD
SEVERELY_SYMPTOMATIC - RECOVERED
ASYMPTOMATIC - RECOVERED

Thinking more about these and playing around with the data, the most useful statistics about these are captured by the output with the compartment values over time. This is what epidemiologists always look at through these SEIR graphs:

I did a small design spin on those and produced the SEIR lines as a heatmap where each compartment is a “row” over the y-axis:

and only looking at symptom related series:

Contacts

contacts.csv which looks like:

time,from,to,weight,label
0,14298,14357,100,Restaurant
0,425,6831,100,School
0,301,15510,100,ConstructionSite

with the following location labels:

array(['Restaurant', 'School', 'ConstructionSite', 'Shop',
       'LargeManyAdultFamily', 'Office', 'CovidHospital', 'LargeAdult',
       'AdultPensioner', 'SmallAdult', 'SmallFamily', 'Hospital',
       'LargeTwoAdultFamily', 'Nursery', 'DoubleOlder', 'CareHome',
       'SingleParent', 'SingleAdult', 'SingleOlder'], dtype=object)  ![](https://paper-attachments.dropbox.com/s_C6E8C0FCB1B16DEE7594A2F1F9B7CE4516FE4FACF347C06F062E142B646BB744_1593701233258_image.png) |
OK, there seems to be some expected numbers here. With school being the most with small family following.

There seems to be 20000 individuals and 8.846.925 contact events in this data.