September 2014 - Darpa-chikun-challenge

CHIKV Challenge Submission #2
by Prithwish Chakraborty 23 Sep '14

23 Sep '14

Hi All, Our next submission is due Oct 1. Back at Arlington, Bryan, Brian and me came up with a few additions to the model mainly 1. Investigating the disease parameters more closely 2. Using a SEIR for base model 3. Constructing a Data driven SEIR to incorporate travel and related information (coding undergoing). Please see the update on that attached herewith. I was wondering whether we can meet sometime tomorrow to share the thoughts on this so that I can finish the coding and get results by Friday. I am proposing either 3-4pm Sep 24, 2014 (tomorrow) or 10-11am Sep 25, 2014 (day after tomorrow). Can you please share your preferences about the same in the following doodle link? http://doodle.com/s4bfat529xpa7qb4 (it has more time-slots) Sorry for being a little late on this update. Thanks Prithwish Chakraborty Graduate Research Assistant Dept. of Computer Science Discovery Analytics Center Virginia Polytechnic Institute and State University email: prithwi(a)vt.edu web: http://people.cs.vt.edu/~prithwi/

1 0

Fwd: DARPA challenge FAQ update
by Prithwish Chakraborty 01 Sep '14

01 Sep '14

Dear Solvers, We have provided these as a response to some common questions. Please note that we have also updated the provided spreadsheet with new data. Delays in PAHO Reports: We recognize there are delays in what PAHO reports and reports by media or local health organizations. However, in order to be fair and consistent in scoring submissions, the final scoring for accuracy and peak prediction will be judged against PAHO reports. The final scores for accuracy of forecast and peak prediction will be based on PAHO data, which will be considered ‘official’ four weeks after the challenge ends, in case there are revisions in PAHO reports. Solvers are encouraged to use all publicly available data sources including news articles and media reports, but should recognize that scoring will be based solely on what PAHO reports, and we cannot validate the accuracy of any information released to the media through non-PAHO channels. Solvers will not be scored for weeks where PAHO data is repeated due to a lag in reporting. However, solvers will be scored for instances where PAHO reports no cases (or the country is not present on the PAHO report for a particular week) and new cases arise during the period of evaluation. Total Cases Decrease between Weeks: In the case that PAHO data reports that the total number of cases decreases between weeks and the cause is not obvious i.e. error in data entry, those weeks will not be scored. Potential Errors in PAHO Reports: 1. Dominica Week 24 & 25: Reported EW 20 cases during Week 25, when Week 24 reported EW 22 data. Week 25 has been set to EW 22 data in the spreadsheet. 2. Sint Maarten Week 5: Reported EW 3 confirmed cases during Week 5. Week 5 has been set to EW 4 data in the spreadsheet. Negative Sign in Accuracy Scoring Formula: The negative sign has been removed in the formula for scoring accuracy. The threshold for eligible submissions remains at 500. This information will also be recorded in the Challenge Resources page. -Kevin InnoCentive Client Services Prithwish Chakraborty Graduate Research Assistant Dept. of Computer Science Discovery Analytics Center Virginia Polytechnic Institute and State University email: prithwi(a)vt.edu web: http://people.cs.vt.edu/~prithwi/

1 0

Re: [Darpa-chikun-challenge] CHIKV Challenge Submission #1
by M Elizabeth Halloran 01 Sep '14

01 Sep '14

Dear Prithwish, Thanks for sending the figures. They predict a lot of Chik cases coming up. This would make sense for a simple SIR model. What are the numbers after the legend where it says "Actual" ? I did not understand a couple of the graphs, for example Saint Martin and Saint Barthelemy. This looks like a great beginning. Best wishes, Betz On 8/31/14, 10:40 PM, Prithwish Chakraborty wrote: > Hi All, > > The predictions results have been pushed to the github repo. These > predictions are from a simple SIR model with no external data. > This baseline method is adapted from Elaine's code with the exception > being different optimizer (constrainedSLSQP > <http://www.pyopt.org/reference/optimizers.slsqp.html>) and using > different scaling around the peaks. > > You can find the figures at > https://github.com/prithwi/dac_chik/blob/master/data/results_sub1/fig/ > and the corresponding csv's in > https://github.com/prithwi/dac_chik/blob/master/data/results_sub1/csv/. > > There are predictions for 20 countries in there. Of the remaining, 30 > don't have sufficient data to fit SIR model. > (see > https://github.com/prithwi/dac_chik/blob/master/data/results_sub1/Notpredic… > ) while for the rest the SIR model gives very poor fit. > > Am attaching the plots herewith for your convenience (as I think many > of the people don't yet have access to the git repo). > I look forward to your comments on the predictions. > > Thanks > > P.S. > 1) Its a very simple model this time mainly due to shortage of time. > But hopefully with our experiences from this time we can add Weather, > Travel Data, etc in the next submission where we will have to predict > the peaks as well. > > 2) Codes have also been pushed to the repo > > Thanks > Prithwish Chakraborty > Graduate Research Assistant > Dept. of Computer Science > Discovery Analytics Center > Virginia Polytechnic Institute and State University > email: prithwi(a)vt.edu <mailto:prithwi@vt.edu> > web: http://people.cs.vt.edu/~prithwi/ > <http://people.cs.vt.edu/%7Eprithwi/> > > > _______________________________________________ > Darpa-chikun-challenge mailing list > Darpa-chikun-challenge(a)cs.vt.edu > https://mailman.cs.vt.edu/mailman/listinfo/darpa-chikun-challenge -- M. Elizabeth Halloran MD MPH DSc Professor Center Stat. & Quant. Infect. Dis. (CSQUID) Vaccine and Infectious Disease Institute Hutchinson Research Center and Department of Biostatistics University of Washington 1100 Fairview Ave N, M2-C200 Seattle, WA98109-1024 USA Phone: 206-667-2722 Fax: 206-667-4378

1 0