youyanggu / covid19_projections Goto Github PK

This is a wonderful tool! I'm especially grateful that you have started modeling a few counties.

I'd love to be able to plot multiple locations in a single plot, say as in the Data Explorer from OurWorldInData. The OWID tool allows one to select what countries to plot, which I find very useful. This tool also changes the URL to match the selected data, so one can go bookmark the data. I confess that I don't know how tough this is :-)

Another (much less widely-used) tool allows one to compare the NY Times' data for states and counties.

I'd love to learn if someone has created an online tool to view your data. If you were to make such plots, I guess your site would be the main place I'd come for COVID-19 data :-)

First, I must say, these projections are really great.

A number of people are looking into excess deaths to determine when exactly the epidemic really started. This project is using official numbers released by governments (which is fine), but maybe another set of projections can be added using excess deaths data. Are there any plans to implement something like this? It would be interesting to see how this affects the projections.

(This is my assumption asking this question: This project uses death data from countries to estimate infection numbers, then predict effective R values, which can be used to generate projections on these numbers. I might be totally wrong about this, correct me if I am. Releasing the code #2 would be helpful to make it clear what approach is taken here.)

Is correction for PPV done, or is it built into the prevalence-ratio/positivity-rate relation? As prevalence drops, tests start becoming less and less predictive for true positivity assuming a fixed sensitivity and specificity.

I noticed that actual_deaths in the US have disproportionately spiked in one day due to yesterday's inclusion of probable deaths by NJ (US_NJ.csv, 2020-06-25).

Here's a news article I found, just to confirm this was NJ's first report of probable deaths in addition to confirmed deaths: https://www.nj.com/coronavirus/2020/06/nj-reports-probable-coronavirus-deaths-for-1st-time-death-toll-now-at-14872-with-170k-total-positive-tests.html

I'm curious if there is a way to incorporate probable deaths without disproportionately throwing them all in a single reporting day? At least from a visual charting perspective, it can be confusing to people. I have no idea if there could be any machine learning projection concerns with it.

The website added a -noreopen option on May 13, such as https://covid19-projections.com/us-noreopen. The projection is very valuable for analyzing the policy effect. I am wondering how can I access the csv files corresponding to those noreopen projection?

I guess the noreopen projection is made based on the historical projection at someday past? So that I can just go back to download the projection at someday past, say May 13, the projection will be corresponding to the noreopen projection? Am I right?

If you got time, it will be great if you could say more about how does the noreopen projection is made comparing to the normal projection?

Not a code issue. Just curious what might explain the massive dip in projected vaccinations in May and June?

Thanks for your great work. I am wondering if you could publish projection in a longer time window, say by the end of this year. I understand that prediction over a very long time might have larger uncertainty. But it is also good to have some information available.

great work!

For all countries, states, and counties it appears that the death predictions are rounded to integers before plotting them or putting them into CSV files. For smaller entities such as Santa Clara CA where the fractional number is between 0 and 1, this creates the awkward case where your predicted total deaths continue to increase while the predicted daily count is 0. So obviously the internal daily death is a fractional number; I think it would be useful to put this fractional number in the plot and CSV file, as you do for the R_t estimates. Of course this extra information may be distracting in the tooltips on plots for large countries or states, yet in that case you can just show the rounded estimate.

Hello,
Your dataset was added to CoronaWhy (https://www.coronawhy.org/) Data Lake on Dataverse as a piece of common COVID-19 data frame http://datasets.coronawhy.org/dataset.xhtml?persistentId=doi:10.5072/FK2/SCH8R0
Would you be willing to help with the maintenance of your dataset in Dataverse, e.g. adding the relevant metadata and keeping the dataset up-to-date? That will help to make the dataset findable and accessible for the medical science community.

Curious about how the uncertainty interval is computed:

That’s why in addition to our most likely estimate, we also provide a 95% confidence interval to reflect this uncertainty. For example, if we predict 150,760 deaths with a range of 88-294k, it means that there is roughly a 95% chance that the true deaths will be between 88-294k.

Here is my guess, hope that you can correct me.

• The uncertainties of the model parameters are first estimated. Some are based on machine learning, such as MCMC? Some are based on assignment, e.g., the uncertainty in fatality rate is assigned as 0.9 - 1.2%.

• Then the uncertainties are propagated to the projection using Monte Carlo, and report the 95% confidence interval.

How is the US current Rt 0.96 as of August 7, 2020?

hi Youyang!

First of all, let me just say that this is marvelous, superbly impressive work. Truly outstanding results, and, imo, orders of magnitude better than the IHME model. I was wondering though, is there any chance of you open sourcing the modeling code? I read your super informative model details page on the site and very much appreciated the explanation there, but I'd still love to see the implementation itself if that's at all possible, though I entirely understand if you choose not do so. Either way, many thanks and much appreciation for your excellent work!

Hi,

Thanks for an amazing website.

Given the importance of the Swedish strategy to COVID19:

Would it be possible to update the model with more accurate death/day from the Swedish CDC?
The John Hopkins numbers are dated based on day reported and not when the person died so lots of spikes due to lack of reporting during weekends/holidays.

This is an Excel file the official Swedish CDC update every day. However there is a 6-7 day lag in reporting so cases currently are only sort of accurate until 16 April.
https://www.arcgis.com/sharing/rest/content/items/b5e7488e117749c19881cce45db13f7e/data

Form this official Swedish CDC site:
https://www.folkhalsomyndigheten.se/smittskydd-beredskap/utbrott/aktuella-utbrott/covid-19/bekraftade-fall-i-sverige/

Current data in JSON:
let sweden = [{"m-d-y":"3/11/20","deaths":1},{"m-d-y":"3/12/20","deaths":0},{"m-d-y":"3/13/20","deaths":1},{"m-d-y":"3/14/20","deaths":1},{"m-d-y":"3/15/20","deaths":2},{"m-d-y":"3/16/20","deaths":2},{"m-d-y":"3/17/20","deaths":1},{"m-d-y":"3/18/20","deaths":6},{"m-d-y":"3/19/20","deaths":7},{"m-d-y":"3/20/20","deaths":9},{"m-d-y":"3/21/20","deaths":8},{"m-d-y":"3/22/20","deaths":12},{"m-d-y":"3/23/20","deaths":11},{"m-d-y":"3/24/20","deaths":20},{"m-d-y":"3/25/20","deaths":23},{"m-d-y":"3/26/20","deaths":31},{"m-d-y":"3/27/20","deaths":32},{"m-d-y":"3/28/20","deaths":35},{"m-d-y":"3/29/20","deaths":39},{"m-d-y":"3/30/20","deaths":43},{"m-d-y":"3/31/20","deaths":47},{"m-d-y":"4/1/20","deaths":52},{"m-d-y":"4/2/20","deaths":69},{"m-d-y":"4/3/20","deaths":78},{"m-d-y":"4/4/20","deaths":71},{"m-d-y":"4/5/20","deaths":86},{"m-d-y":"4/6/20","deaths":91},{"m-d-y":"4/7/20","deaths":83},{"m-d-y":"4/8/20","deaths":111},{"m-d-y":"4/9/20","deaths":84},{"m-d-y":"4/10/20","deaths":89},{"m-d-y":"4/11/20","deaths":96},{"m-d-y":"4/12/20","deaths":95},{"m-d-y":"4/13/20","deaths":84},{"m-d-y":"4/14/20","deaths":89},{"m-d-y":"4/15/20","deaths":102},{"m-d-y":"4/16/20","deaths":99}]

Hello, I've read through all of your documentation and that information you're compiling and putting together is impressive. Do you have an API? I'm interested in comparing your data to some of the other models.

Best,
Taylor

I believe this was already asked, I was wondering if this has changed at all? As I would like obviously publish a longer prediction with some methods I had in mind.

Thank you,
Montana.