New projection model

[ajkeith]

@PriyaRutgers, thanks for joining the team! I'm making this issue so we have a place to discuss any roadblocks.

[ajkeith]

This is where load model data: https://github.com/CHAD-Analytics/CHAD/tree/master/covid19/www/4_load_external_data.

This is where we build the interface: https://gitlab.devforce.disa.mil/A9-ADV/chad/-/blob/integration/CHAD/covid19/ui.R.

[PriyaRutgers]

I went through the interface thoroughly, spotted few maps overlapped with tables, and few such UI issues. Also for certain locations the lower, upper bounds and other info for the prediction models is not displayed. Also the updated IHME models have incorporated the interventions ( relaxing stay at home orders, mask mandates etc. as states have started to reopen), do you suggest updating the model that is currently displayed on the dashboard?

Let me know, I will get started. Thanks!

[tleonard6]

Are you working from the Master? If so, then please feel free to jump on any fixes or errors. Also, updated the IHME model would be great. Do you have a summary of the mode that you had been working on previously? Thank you!

…

On Sun, Jun 28, 2020 at 11:49 PM Debopriya Ghosh, PhD < ***@***.***> wrote: I went through the interface thoroughly, spotted few maps overlapped with tables, and few such UI issues. Also for certain locations the lower, upper bounds and other info for the prediction models is not displayed. Also the updated IHME models have incorporated the interventions ( relaxing stay at home orders, mask mandates etc. as states have started to reopen), do you suggest updating the model that is currently displayed on the dashboard? Let me know, I will get started. Thanks! — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#51 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AG6TX46QFDXTHRIMC4GLHRTRZAFMHANCNFSM4OBVMDKQ> .

[PriyaRutgers]

Sure I can start with fixing those. I was referring to the ARGO Models (reference below)-
https://www.nature.com/articles/s41467-018-08082-0.pdf
https://www.pnas.org/content/112/47/14473

For our model, we are using the world meter data instead of CDC data because of the lag in CDC reporting. And for exogenous variables we are incorporating lot of big data -- twitter feeds (maintained by Johns Hopkins, healthtweet.org), Google trends related to health, consumer products, and technology related search terms mostly. In addition we are using a clustering approach to identify counties that have same demographics, and socio-economic structure. Compared to the papers I mentioned above we performed some further fine tunings for the search terms as Google Correlate is no longer available and so to find the pattern in these search trends more accurately we used an approach called spectrogram. We found the time series for counties in same cluster are correlated and tested predictive value using Granger Causality test. We have gathered data for NY, NJ, CA and FL so far. I can share with you some results (plots and statistical measures) this Tuesday. Our next step is to perform change point analysis (Bayesian approach vs our proposed approach) on these time series to identify how they relate to the mitigation strategies incorporated.