SERVING THE QUANTITATIVE FINANCE COMMUNITY

  • 1
  • 2
  • 3
  • 4
  • 5
  • 31
 
User avatar
trackstar
Posts: 27647
Joined: August 28th, 2008, 1:53 pm

Re: Models for Covid-19

March 22nd, 2020, 7:42 am

COVID-19 Open Research Data Set


https://pages.semanticscholar.org/coronavirus-research

“ In response to the COVID-19 pandemic, the Allen Institute for AI has partnered with leading research groups to prepare and distribute the COVID-19 Open Research Dataset (CORD-19), a free resource of over 44,000 scholarly articles, including over 29,000 with full text, about COVID-19 and the coronavirus family of viruses for use by the global research community.

This dataset is intended to mobilize researchers to apply recent advances in natural language processing to generate new insights in support of the fight against this infectious disease. The corpus will be updated weekly as new research is published in peer-reviewed publications and archival services like bioRxiv, medRxiv, and others.”

A very small subset of specific article recommendations will be provided in the morning.
Last edited by trackstar on March 22nd, 2020, 10:23 pm, edited 1 time in total.
 
User avatar
Cuchulainn
Topic Author
Posts: 63413
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Re: Models for Covid-19

March 22nd, 2020, 11:22 am

Since the research is not open (it seems), we can only guess what the implicit assumptions,use cases and methods are. It looks like the methods used are based on standard applied mathematics for ODEs and in many cases they attempt to find quasi-analytic solutions?

There is more to life than ODEs but my question is why these reports do not discuss these problems as dynamical systems ("ODEs++") so that other questions can be answered beyond numerical ODEs. I wonder  why no mention is made of the Lyapunov, Poincare-Bendixson theory etc. What happens to the ODEs when [$]t \rightarrow \infty[$] (sounds like a logical question). A Sapir-Whorfe problem ("give hammer then all is a nail")

What's missing IMO is what the scope of the research is.

A good start is to publish the precise details of that 2-minute "blind me with science" piece on BBC News. It came across as a rare piece of voodoo.
My C++ Boost code gives
262537412640768743.999999999999250072597198185688879353856337336990862707537410378210647910118607313

http://www.datasimfinancial.com
http://www.datasim.nl
 
User avatar
Cuchulainn
Topic Author
Posts: 63413
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Re: Models for Covid-19

March 23rd, 2020, 12:04 am

YIKES, 13 years ago. And it hasn't been updated since?
I don't want to see your code, just the maths model. A wild guess you can do in [2K, 3K]LOC with C++11.

Is the C code an implementation of a generic MSEIR model is taken out of mothballs for COVID-19? I wonder about the 'engineering quality approach'.. How many people work(ed) on the product and how mature is it?

I'm speechless, Ugh. As a software product/process it looks primitive. And  the BBC took it as Gospel last week. They blinded us with science.
Last edited by Cuchulainn on March 23rd, 2020, 12:26 am, edited 2 times in total.
My C++ Boost code gives
262537412640768743.999999999999250072597198185688879353856337336990862707537410378210647910118607313

http://www.datasimfinancial.com
http://www.datasim.nl
 
User avatar
Paul
Posts: 10922
Joined: July 20th, 2001, 3:28 pm

Re: Models for Covid-19

March 23rd, 2020, 12:23 am

Not necessarily bad that it hasn’t been updated.

Bad that it’s seven dependent variables.

Bad that the code is so long and undocumented. And why thousands of lines for seven odes?
 
User avatar
Cuchulainn
Topic Author
Posts: 63413
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Re: Models for Covid-19

March 23rd, 2020, 12:32 am

Paul:
Not necessarily bad that it hasn’t been updated.

Bad that it’s seven dependent variables.

Bad that the code is so long and undocumented. And why thousands of lines for seven odes?
In C++ these 'magnificent 7' ODEs - neglecting the data input/output parameter parts - can take about [20,70] lines using an ODE solver and a few days work at most. Here is Anchor code to give you an idea. (7 is too much indeed).

dxdt(i, j) = A * (tmpP - 2 * tmpMid + tmpM)
		+ D * (tmpP - tmpM)
		+ sgn * E*(U(i, j + sgn) - tmpMid)
		+ F * tmpMid
		+ G;

My C++ Boost code gives
262537412640768743.999999999999250072597198185688879353856337336990862707537410378210647910118607313

http://www.datasimfinancial.com
http://www.datasim.nl
 
User avatar
zeta
Posts: 1973
Joined: September 27th, 2005, 3:25 pm
Location: Houston, TX
Contact:

Re: Models for Covid-19

March 23rd, 2020, 2:28 pm

Thanks for sharing trackstar. Time permitting I might be able to lend some hpc/machine learning/numerical chops. A quick parse of the text did show that some researchers suspect the faecal-oral transmission route, which reminds me of rotavirus, something that's treatable now but is still ubiquitous/pretty devastating in the developing world. 
 
User avatar
Cuchulainn
Topic Author
Posts: 63413
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Re: Models for Covid-19

March 23rd, 2020, 2:44 pm

I am happy to say that @Microsoft and @GitHub are working with @Imperial_JIDEA and @MRC_Outbreak  to document, refactor and extend the code to allow others to use without the multiple days training it would currently require (and which we don’t have time to give)...
- Neil Ferguson

aka? a) don't call us, we'll call you, b) we'll get back to you?

BTW is the IC model the only kid on the block and why the meteoric rise in the media? 
My C++ Boost code gives
262537412640768743.999999999999250072597198185688879353856337336990862707537410378210647910118607313

http://www.datasimfinancial.com
http://www.datasim.nl
 
User avatar
trackstar
Posts: 27647
Joined: August 28th, 2008, 1:53 pm

Re: Models for Covid-19

March 23rd, 2020, 2:45 pm

@zeta
Interesting observation and good to have the kind of expertise that can be found around here. 
I am working on finance and economics today, but will come back to the Covid research too, as you say, time permitting.
(Still reviewing the literature and working on selecting the best articles.) 
It's been an avalanche, but some stand out. I will post some later this week.
Last edited by trackstar on March 23rd, 2020, 3:13 pm, edited 1 time in total.
 
User avatar
trackstar
Posts: 27647
Joined: August 28th, 2008, 1:53 pm

Re: Models for Covid-19

March 23rd, 2020, 2:48 pm

I am happy to say that @Microsoft and @GitHub are working with @Imperial_JIDEA and @MRC_Outbreak  to document, refactor and extend the code to allow others to use without the multiple days training it would currently require (and which we don’t have time to give)...
- Neil Ferguson

aka? a) don't call us, we'll call you, b) we'll get back to you?

BTW is the IC model the only kid on the block and why the meteoric rise in the media? 
Let's put that in context - this is from Neil Ferguson's Twitter feed, where he announced yesterday that there has been tremendous interest in their model and they are working with big tech partners (as you see) to make the model and code available to others.

I think the best thing now is to watch that feed and also press releases from IC JIDEA - it will certainly be announced when ready.

I don't think it is "the only kid on the block," but this model made its way to policy-makers here (and elsewhere) and got traction. Then mentioned as a key influencer on decision-making processes and mainstream media grabbed it. Neil is fairly well-known in the outside world, at least among certain elites, so that is part of it too. At least that is how it seems from here.

Some useful links: 
https://twitter.com/neil_ferguson
https://twitter.com/Imperial_JIDEA
https://www.imperial.ac.uk/jameel-insti ... formation/
Last edited by trackstar on March 23rd, 2020, 2:56 pm, edited 2 times in total.
 
User avatar
Cuchulainn
Topic Author
Posts: 63413
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Re: Models for Covid-19

March 23rd, 2020, 2:49 pm

Thanks for sharing trackstar. Time permitting I might be able to lend some hpc/machine learning/numerical chops. A quick parse of the text did show that some researchers suspect the faecal-oral transmission route, which reminds me of rotavirus, something that's treatable now but is still ubiquitous/pretty devastating in the developing world. 
What use cases could be tested? BTW the metadata is in EXCEL  :shock:

I was always led to believe that ML/ANN only works when there is enough training data? So, how does it work for Covid-19?
My C++ Boost code gives
262537412640768743.999999999999250072597198185688879353856337336990862707537410378210647910118607313

http://www.datasimfinancial.com
http://www.datasim.nl
 
User avatar
trackstar
Posts: 27647
Joined: August 28th, 2008, 1:53 pm

Re: Models for Covid-19

March 23rd, 2020, 3:16 pm

Perhaps use SARS (with caution) as a similar case and make adjustments?

Also able to produce individual models using the cruise ships, China as longest duration study and includes a tight lockdown, and Washington state as a case where it enters into a frail, elderly population in institutional setting almost immediately upon arrival (traveller).

Back to the Johns Hopkins global data set and then question - how good is it? And how can it be supplemented with local data that may be more accurate or have nuances not apparent in the big picture.

NY is moving fast here now; has vaulted over Washington. But that is not a surprise.

Mapping the spread of coronavirus in the US and worldwide - The Washington Post March 23 update
 
User avatar
Cuchulainn
Topic Author
Posts: 63413
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Re: Models for Covid-19

March 23rd, 2020, 4:30 pm

OK then, I am now setting up the MSEIR ODE model in C++ as here, I hope to get back in a few days. Since I am not a epidemologist, I will need your support to tell  what the ODE parameters are and their values (e.g. for chickenpox, the infectious period [$]1/\gamma[$] is 6 months, etc.) 

Two major use cases are [$]\displaystyle \lim_{t\to\infty}s(t)[$] and [$]\displaystyle \lim_{t\to\infty}i(t)[$] for susceptible and infectious fractions of the population at the end of the epidemic. And computing the contact number [$]\sigma[$]. 

// If age is an independent variable then you get a system of 1st order hyperbolic PDEs, a topic I did my MSc research  on using FEM some 45 years ago.
My C++ Boost code gives
262537412640768743.999999999999250072597198185688879353856337336990862707537410378210647910118607313

http://www.datasimfinancial.com
http://www.datasim.nl
 
User avatar
Cuchulainn
Topic Author
Posts: 63413
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Re: Models for Covid-19

March 23rd, 2020, 4:36 pm

ImageImage
My C++ Boost code gives
262537412640768743.999999999999250072597198185688879353856337336990862707537410378210647910118607313

http://www.datasimfinancial.com
http://www.datasim.nl
 
User avatar
Paul
Posts: 10922
Joined: July 20th, 2001, 3:28 pm

Re: Models for Covid-19

March 23rd, 2020, 4:40 pm

Is that the model?
 
User avatar
trackstar
Posts: 27647
Joined: August 28th, 2008, 1:53 pm

Re: Models for Covid-19

March 23rd, 2020, 4:43 pm

OK then, I am now setting up the MSEIR ODE model in C++ as here, I hope to get back in a few days. Since I am not a epidemologist, I will need your support to tell  what the ODE parameters are and their values (e.g. for chickenpox, the infectious period [$]1/\gamma[$] is 6 months, etc.) 
From the How should we threading OT - a paper I posted last week - noticed you have mentioned Cascella here earlier - he is also a coauthor on this one:

Here is a recent paper on the NCBI website:

Features, Evaluation and Treatment Coronavirus (COVID-19) Marco Cascella et al. - update March 8 2020

Current view (at least here) is that Ris 2.2 - which is in the range cited elsewhere over the past few weeks - 2.0 - 2.5 from what I have seen.

"Based on data from the first cases in Wuhan and investigations conducted by the China CDC and local CDCs, the incubation time could be generally within 3 to 7 days and up to 2 weeks as the longest time from infection to symptoms was 12.5 days (95% CI, 9.2 to 18).[5] This data also showed that this novel epidemic doubled about every seven days, whereas the basic reproduction number (R0 - R naught) is 2.2. In other words, on average, each patient transmits the infection to an additional 2.2 individuals. Of note, estimations of the R0 of the SARS-CoV epidemic in 2002-2003 were approximately 3.[6] "

I have also read papers that stated it could continue to be contagious even after the person has "recovered" - still showing some positives about two weeks later. Clearly a problem if so, with the recovered folks getting back to their normal lives. 


Is that part of the 1-2 punch?

Even so, these self quarantines and shut downs should help if they can last 2 weeks or "ideally" a month, perhaps.


/end quote

NY at 16,916 now, change is +44% from Saturday

Italy at 46,638, change is +9% from Saturday
Last edited by trackstar on March 23rd, 2020, 4:48 pm, edited 1 time in total.
ABOUT WILMOTT

PW by JB

Wilmott.com has been "Serving the Quantitative Finance Community" since 2001. Continued...


Twitter LinkedIn Instagram

JOBS BOARD

JOBS BOARD

Looking for a quant job, risk, algo trading,...? Browse jobs here...


GZIP: On