appliedbinf / covid19-event-risk-planner Goto Github PK

• disable caching for the map HTML assets

Brief Overview of Issue

When attempting to run this project in a Docker container and access it (via port forward 3838), the page is greyed out and the cursor is disabled. Pulling up the web browser js console, I see the following output:

The application unexpectedly exited.

Diagnostic information is private. Please ask your system admin for permission if you need to check the R logs.
[INF]: Connection closed. Info: {"type":"close","code":4503,"reason":"The application unexpectedly exited","wasClean":true}

I had to comment out a few lines in the Dockerfile to complete the build and run my container. The following lines were commented out of the Dockerfile:

COPY .rtweet_token.rds /root/.rtweet_token.rds
...
COPY docker_github /root/.ssh/id_rsa
COPY docker_github.pub /root/.ssh/id_rsa.pub
RUN ssh-keyscan -H github.com >> /root/.ssh/known_hosts
...
RUN git clone [email protected]:appliedbinf/covid19-event-risk-planner.git /root/repo

My Question

I saw in a previous issue that this application was dockerized for deployment. Are there any plans to make this application run in a docker container for development purposes?

the USA Risk Estimates By County map and the Global Risk Estimates map won't load on the site. not sure if this is a hosting issue. just a quick heads up.the continouous risk estimates and real time estimates is still functional.

Add the risk calculator to the main maps page

Visual tweaks to calculator:

• Grid lines
• Placeholder values
• Check invalid inputs (highlight on invalid)

Iowa's API is causing problems for the NYTimes, so the NYTimes data is not refreshing properly for Iowa county-level data. It looks like the case numbers aren't changing, so the Iowa numbers are artificially low.

It's obvious now that the risk is 0, but was misleading when there was still good 2-week old data left.

I understand doing something about it opens an entire can of unpleasant worms, alas.

https://github.com/nytimes/covid-19-data/blob/master/us-states.csv

The ascertainment bias is a bit outdated for most places - it would be fabulous to have a slider with a few options

My understanding from looking at #6 is that you are pre-rendering the maps, so this could substantially add to that, but perhaps just allowing for 3 options (1:10, 1:5, 1:2) would suffice?

UK is reporting cases in City of London (E09000001) and in Hackney (E09000012) together. Similarly Cornwall (E06000052) and the Isles of Scilly (E06000053) are reported together.
Suggest merging spatial polygons, names (e.g. Hackney and City of London), and population sizes for these areas.

Hi,
Not sure if this is the right place to raise this, but my understanding of the method being applied here is to assess risk as a function of state/county-level cases / state/count-level population. Is there any adjustment being done for the fact that events are somewhat unlikely to just draw people from the same state and very unlikely to just draw people from the same county? Looking at the map for 1k-person events, I see urban areas with vastly different risk estimates even though an event with 1k people is very likely to draw participants from the surrounding area.

Maybe I'm missing it in the code. It seems like there should be a spatial/density component here.

I'm trying to use this model to create a tool for theatres in our county to use to work backwards and determine safe values for reopening (determine target caseload based on permissible risk), but I'm unable to precisely recreate your model. I'm using the State of PA right now - the data cited lists a population of 12,801,989 and reports current active caseload at 24,574 (from the Real-Time State Level estimate page). Running the Calc_Risk function on those values for an event of 500, at 1x Ci:
p_I <- I / pop
1 - (1 - p_I)**n
becomes
1-(1-(24574/12801989))**500
or 61%.

However, the web tool reports 49.6%, which (working backwards) assumes a caseload of 24,574 for a population of 17953987.
https://i.imgur.com/lOkBeDe.png
So I think the tool is misreporting either active caseload or risk, or using an incorrect population assumption.

If I run the formula again with that incorrect population 17953987 for other scenarios (i.e. event of 50 with at 5x Ci) the results I get match perfectly.
https://i.imgur.com/2P5yp52.png

Apologies if I'm completely missing something, but I'm just concerned that I can't recreate the results from the tool on my own. I'm not an R programmer myself, so I may simply be misinterpreting something in your code.

Hi, are the daily/current datasets used by the front end at https://covid19risk.biosci.gatech.edu available? I was hoping to avoid hosting my own server.

The paper is out in NHB! Update the in-app citation: https://www.nature.com/articles/s41562-020-01000-9

To news section:

• https://www.webmd.com/lung/news/20201109/whats-my-risk-of-covid

Add link to Spanish site: https://eventosycovid19.es/

Lower left corner of page presents data for the 10 day average Ci, while the plot displays data for the 14 day average Ci.
Example: 11/16/20, California, Event size 10, Ci (88,717)
Lower left corner Plot
Chance someone is COVID19 positive at CI : 1.6% 2.22%
Chance someone is COVID19 positive at 5xCI : 7.7% 10.68%
Chance someone is COVID19 positive at 10xCI : 14.9% 20.32%

The risk level (%) legend in the USA estimates tab appears to be missing from the global estimates tab. I think it would be useful to add this.

Note that the structure of the data resource for Czechia has changed (I think) this week. The file now includes a new column for id as the first column in the dataset. Could amend:

as

names(czechData) <- c('id','Date','code','District','Confirmed','Cure','Death')

Tabs load, map does not. Seems to work in Chromium based browsers.

TBD

I've been posting time series plots of risk for the D.C. metro area (D.C. + northern VA + parts of MD). I get lots of questions. Most of them are variants of the following:

Me: In a gathering of 10 people, there’s a 15–20% probability that at least 1 person is infected.

Them: Wouldn’t that mean that 10% of the total population currently has a COVID infection?

To which my response is:

No.
If the probability is 100%, one or more people are infected---there is no doubt. If the probability is 0%, no person is infected---there is no doubt. If the probability is 20%, you are more certain that one or more people are infected than if the probability is 10%.

I get so many questions like this that I'm no longer going to use percentages.

Sometimes there's a follow-up comment along the lines of:

Them: That doesn't help me, because I don't know where to set the risk I'm willing to personally accept.

To which my response is:

Me: Say you're willing to accept a risk of 0.17. Flip a coin 10 times. The risk of getting 7 or more heads is 0.17.

A better second response has been hard for me to come up with, but a coin flip is something people know.

Food for thought.

Truncate the timestamp and move the % to no-space

For both tweets and site, testing is slowing down

Add blog post to about page and press

• Limit direct push activities to master
• Require changes to master to go through PRs
• Enforce code review on PRs (allow self reviews)
• Disallow non-linear commits and FF
• Turn restrictions on for admins too (aka me)

The continuous risk estimates have a way to sort by rank, and they also have a way to sort by rank on a particular day. Would it be possible to add a chart showing the rank over time, in order to see the changes in a particular state or all states? If not rank, then %chance over time, perhaps at a hardcoded or selected event size?

• data_cur date: cases
• data_past date: cases
• population

e.g.

2020-10-28: 250 cases
2020-10-14: 220 cases
population: 8500

Otherwise, it's impossible to infer how risk was calculated.

First, thanks for the code and making this project open. I'm using this for my effort to build a school re-opening calculator. Community prevalence is a key parameter.

With that said, I want to share my investigation into Ascertainment Bias.

Based on the CDC's October Seroprevalence data, I think the Ascertainment Bias for new cases/active cases is closer to 3.0.

I've written this up here:
https://github.com/chonghorizons/covid-estimators/wiki/Ascertainment-Bias

OPTIONS

0. Do nothing.
1. You might allow people to use 3.0 in the main chart.
2. You might default to 3.0 or 5.0.
3. (hardest) Pull in the state-by-state ascertainment bias numbers.

• For NY/NJ data, make adjustments for AscertainmentTotal (Total Infected/Total Tested Positive) and AscertainmentActive (ActiveInfected/ActiveTestedPositive)
• The seroprevalence data has high uncertainty, so you can't really do a first-differencing of the seroprevalence data waves.
• You might be willing to use a neighboring-state number. Or try to fill in information for some states using test-positivity as a regressor.

Again, thanks!!!

nytimes/covid-19-data#447

Sept. 2: Massachusetts revised its methodology for probable cases and deaths, removing 8,050 previously announced cases and 26 deaths.

Add more options at the low end, make 500 (?) the top end

Update docs

Should probably include sections on:

• Development
  • With Docker
  • Without Docker
• Building image
• First start / required flat files
• Deploying

Other misc:

• Clean up and check in Compose file
• Provide stubs for purposefully not provided files (ie. rtweet tokens)

Move to ragg for substantially faster rastering, especially for state plot grids

Radio button on-off: "Show me typically thanksgiving trips"

Network of known common trips/travel, e.g.

Yesterday's code update changes ui.R to include www/mt.html, but the file isn't tracked in github.

Work with @seolhalee to integrate new EU maps into site and twitter

• Remove some less used tabs
• New EU specific tab

When drilling down into the map to view county-level results, navigation is already somewhat awkward (slow rendering, two-finger pan gesture on Mac affects zoom instead). This is further complicated in that changing the event size will reset the map to full overview. This makes it more challenging than necessary to obtain county-leven risk for different event sizes, or just explore the visualization.

Thanks for the good work!

• Update VM geospacial libraries (missing dependency)

• Integrate Seolha's code into a new hourly (?) script

• Add htmlwidgets to app

i suggest adding some text prompts advising viewers to disable adblock. the newer features and overlays will prevent the site from working if adblock is turned on in the web browser.