Profile picture
, 7 tweets, 2 min read
My Authors
Read all threads
Today I am open-sourcing the SEIR simulator behind the YYG / covid19-projections.com model.

If your system supports Python, you can generate your own simulations in under 5 minutes.

github.com/youyanggu/yyg-…
This is the underlying SEIR model without the machine learning layer to learn the parameters. We do include "best" parameter sets learned by the full model.

This simulator does not use any published data. As a result, this tool is meant to generate simulations, not projections.
Because this simulator has very little dependencies (only NumPy), it is fast to run, works right out of the box, and is easily modifiable.

I've purposefully designed the simulator to be as lean and simple as possible (since all of the computation is run on my laptop).
Our full model simply runs thousands of parameter sets through this simulator. It then compares the resulting simulations with the observed data, and combines the most accurate simulations to generate a single projection.
You can use this simulator to create interesting scenarios like:

- How much can we reduce infections/deaths if people started social distancing x days earlier
- What if y% of individuals self-quarantine after symptom onset
- What if reopening was pushed back by z days
Thank you to all for being patient - it took a while to find the time to put this together. I am committed to open and transparent research.

I'll try my best to answer any questions. Apologies in advance if I don't get to everyone.
I also encourage everyone to also read our Model Details and Assumptions pages to get a better high-level idea before diving into the source code:

covid19-projections.com/model-details/
covid19-projections.com/about/#assumpt…
Missing some Tweet in this thread? You can try to force a refresh.

Keep Current with Youyang Gu

Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

Twitter may remove this content at anytime, convert it as a PDF, save and print for later use!

Try unrolling a thread yourself!

how to unroll video

1) Follow Thread Reader App on Twitter so you can easily mention us!

2) Go to a Twitter thread (series of Tweets by the same owner) and mention us with a keyword "unroll" @threadreaderapp unroll

You can practice here first or read more on our help page!

More from @youyanggu see all

Profile picture
Youyang Gu
@youyanggu
Every week, I along with other researchers/modelers gather in a conference call to discuss the current state of forecasting, give presentations, and share ideas.

One group always absent? IHME.

They are not representative of the scientific community.
While the senior leadership at IHME declines to participate in these collaborative efforts, they frequently have time to appear on national television, give press conferences, and speak to the media.

It's clear where their priorities lie.
Because of IHME's massive media influence, their past projections have had detrimental consequences.

Experts had been projecting 100-240k deaths since March. Instead, throughout April, all we've heard was "influential model revises projections to 60-70k deaths by August".
Read 9 tweets
Profile picture
Youyang Gu
@youyanggu
I created a map that shows how cases are changing in each state normalized by both population and rate of increase.

Raw changes in cases can be misrepresentative, so I created a score that tries to encapsulate a fuller picture.

Full interactive version: covid19-projections.com/maps/
Based on this calculation, Arizona has the lowest score based on its change of +69 cases per day/1 million and 98% increase compared to 7 days ago.

Other states of concern are: TN, AR, NM, FL, UT.

On the other hand, Mid-Atlantic states continue to see a steady decline in cases.
This map slightly deviates from my traditional analysis in that this map does not incorporate any projections.

These are simply a visualization of the raw case data from @JHUSystems.
Read 5 tweets
Profile picture
Youyang Gu
@youyanggu
We have lowered our projections 15% from its peak 2 weeks ago due to recent positive data showing a continued steep decrease in infections and deaths in many states.

Furthermore, states that have reopened have reopened slowly, reducing the likelihood of an immediate second wave.
The above graph shows all states with >5% YOY diners on May 10. All 8 states still have <50% YOY diners a month after the reopening.

Of course, there are capacity limits in place, but the effect is likely minimal since most restaurants are not constantly at full capacity.
I also think many of the deaths reported in early May were April deaths that were backlogged/retroactively added, hence skewing the data.

As a corollary, we have lowered the IFR for the US from a constant 1% to 0.75%. This is to reflect improving treatments and age distribution.
Read 5 tweets

Embed code for your website

×
Follow Us on Twitter!

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3.00/month or $30.00/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!