C19 Update 18 Oct 2022

It has been a while since we did a covid update for AR, US and World. But here it is 2 years and 10 months into the covid era. 

 

C19 Update

C19 Update

 

C19 Update

C19 Update

 Lordy!

C19 Update

 

C19 Update

 Cases continue extremely high and setting 7-day average records in all zones plotted. I expect the usual 2 week lag to see the signal in the death data.

C19 Update - Something Strange

Below is today's AR-US-World plot of C19 deaths (left) and cases (right). Note US and world case data for today are crazy high. In fact, US is over 1M daily cases, representing 40% of cases reported worldwide. The AR 7-day average cases is the highest ever and steeply increasing. Meanwhile deaths for all regions are trending steady or down. Are we seeing the omicron effect -- more infectious and less deadly?

As the song says, 'Something's happenin' here, what it is ain't exactly clear."

C19 Update

 

C19 Update

 

Fall 2020 vs Fall 2021

I could write a novella interpreting these graphs, but they pretty much speak for themselves. Large open circles show beginning of fall 2020 and fall 2021 semesters. We started last fall face-to-face teaching with 'covid spacing' for students, masks recommended. This fall we are face-to-face, full density, masks required. 

 

C19 Update

 

C19 Update

 

C19 Update

 

India Questions and Fit

 This note came from reader CM in India. 

I just looked at your SEISMOS predictions for US deaths data.

There are two points I do not understand:

1. Why is there so much vertical scatter in the observed data? 

2. Your predications are based on data before dec 2020. What about predictions based on the latest available data? Would that not be more meaningful?

Also, can please provide predictions based on Indian data? There are several speculations about the oncoming "Third wave" and it would be very useful to have a reliable prediction. 

Thank you for reaching out CM. 

The US data scatter is due to uneven reporting on weekends vs weekdays, an effect that is much less evident in India data. Congratulations!

My point in making predictions based on US data through 19 Dec 2020 was to estimate, however crudely, the lives saved by vaccination since that date.

You asked for a fit of India C19 death data. As requested, result below. The outlying points may have made the fitting difficult, it took 37 iterations to get accuracy below 500 on the cumulative curve. I see no indication of a third wave.

C19 Prediction Check from 12/19/2020

 Back on 19 Dec 2020 I posted a fit of C19 US death data based on three lognormal curves, a total of 12 free parameters. The date is significant: Pfizer vaccine received emergency approval by the FDA on 10 Dec and Moderna the 19th. So all the data before 19 Dec was pre-vaccine. I archived the fit parameters from that day to allow comparison with actual death data that unfolded in the coming months after introduction of the vaccines. 

It has been 200 days since I fit the 19 Dec data. The plot below shows how well, or not well, the prediction worked. The top plot is cumulative US C19 deaths, with data before 12/19/2020 in blue dots and red dots for later data. The blue dashed line is the prediction based on the blue points only. In the lower plot, US covid daily death data are shown with the same color scheme. We can never know if the un-vaccinated pandemic in the US would have followed my prediction, but if we believe it was on the right track then vaccines have already saved 650K lives in the US alone. 

II hope you bought some Moderna stock and made some money, it is vegan you know. The real value is not on Wall Street, but in the many lives that carry on. Sometimes a normal life, just normal, is a beautiful thing. 

C19 Update - India (again) and Source Code

CM in India asks: "Can you pl share the outline of your LG3 algorithm that you used for prediction?" 

CH, I have put the python source code in github at

https://github.com/cliner21/fit_Covid_LG3

----- GitHub readme file -----

Region fit of Johns Hopkins Covid-19 data using three lognormal functions. This code is open for public use.

Python3 code with several standard library includes. Assumes c19subs.py is in the same directory. The code is a bit sensitive. The fit is across a 12-dimensional parameter space and a good starting point (p0) and set of parameter bounds are essential. The bounds and p0 are generated randomly for each iteration. The core routine is scipy.optomize set with maxfev=5000, but if p0 or bounds are pathelogical optomize may run to full 5000 function evals without convergence and code will terminate with error. Rare event, just run it again and hope for the best.

The user needs to supply an accuracy goal, this is set to target=500 which seems suitable for India as of 9 May 2021.

To run code from the command line:

python fitLG3region.py

With (a) result:

searching: iter= 1  goal= 500.0  accuracy= 50600
searching: iter= 2  goal= 500.0  accuracy= 10000000
searching: iter= 3  goal= 500.0  accuracy= 90499345
.
.
.
searching: iter= 34  goal= 500.0  accuracy= 120131516
searching: iter= 35  goal= 500.0  accuracy= 357248
searching: iter= 36  goal= 500.0  accuracy= 463

The generated fit for this run is in the file LG3_May09_India.pdf

Author: Christopher Liner

-----

C19 Update India

From CM in India: As you know India is totally gripped by the second wave of Corona. People are speculating about when the third wave will hit us and how deadly it will be. I shall be personally thankful to you if you can run your ANN prediction for India data.

Dear CM: I ran the best neural network I have come up with to no avail. With an accuracy target of 2500 it took 49 iterations and still did not fit the data in any meaningful way. As you probably know, over the course of C19 my work on neural nets only managed to confirm what experienced NN workers have long known -- neural nets are not very good at predicting outside of the given data. So the NN was a flop. But, I did manage to get a decent fit on my LG3 model. This is a sum of three lognormal functions with 12 free parameters used to fit the data. Like all data fitting approaches, the key is choice of starting parameter values. Anyway, here is the India C19 deaths plot fit to my LG3 model. On the current steep trajectory, we can expect the predicted peak location and value to change rapidly with new data points. Even so, it looks like difficult times ahead for my many friends in India. 

C19 Update

 It has been a while, but here is the current AR-US-World plot. The world is not looking good.

C19 Update

 The virus year is almost over, in the US at least. I plan to make a last update on 8 Apr 2021, exactly one year after my first covid-19 Seismos post of 8 Apr 2020. It is time to move on to other things.