Lockdown number crunching with the HP Prime

[JimP]

How is Melbourne tracking under its roadmap to reopening? The Victorian Premier has stated a supercomputer was used to determine scenarios for the Sept 28, Oct 26, Nov 23 dates based on various parameters. Using an HP Prime's 2Var statistics and Function apps, one can show the almost perfect logistic curve fitting equation based on the official data can be used to predict that for all other things being equal (ie the Stage 4 lockdown) the state will reach 50 cases (14 day moving average) on 16 September, 30 cases (14 day moving average) by 23 September (this is the whole state, not only Metro Melbourne) and <5 cases (14 day moving average) by 15 October. On that latter date the number of active cases should also have dropped below 200 based on a similar logistic curve analysis. To get to <1 new case (14 day moving average) -- the curve will never reach zero -- that will take until 3 November. But to reach <1 active cases in the state -- that will take until mid January 2021. Realistic? Let's see how well supercomputers vs HP Prime can do...

[Paul Dale]

When the numbers get small, individual probabilities will start to impact. One person infecting their family or a single missed contact trace e.g. When the numbers are large, the data will fit well.

[Albert Chan]

Fitting historical data is almost always perfect. (If not, just add more variables)
Predicting the future (i.e extrapolation) ... not so much.

Of course, with so many predictions proposed, some will hit it bull's eye

Nasssim Nicholas Taleb's book, "The Black Swan", page 146

Quote:The bookmakers were given the ten most useful variables, then asked to predict the outcome of [horse] races.
Then they were given ten more and asked to predict again.

The increase in the information set did not lead to an increase in their accuracy;
their confidence in their choices, on the other hand, went up markedly.

Information proved to be toxic.

These may be of interest:
https://plus.maths.org/content/covid-19
https://statisticsbyjim.com/regression/r...-too-high/
https://www.barrons.com/articles/chinas-...1581622840

[Pjwum]

Your extrapolation algorithms should make use of the SEIR model. I do it in Excel, but parameters change almost weekly.

You can do it even online very comfortably and with nice animation.

[Didier Lachieze]

(09-11-2020 11:52 AM)Albert Chan Wrote:  Fitting historical data is almost always perfect. (If not, just add more variables)
Predicting the future (i.e extrapolation) ... not so much.

It' like weather forecasting. In french we say: "la météo est la science qui prédit le temps qu'il aurait du faire", which translate to something like "Weather forecasting is the science that predicts the weather as it should have been"

[JimP]

(09-11-2020 03:59 PM)Pjwum Wrote:  Your extrapolation algorithms should make use of the SEIR model. I do it in Excel, but parameters change almost weekly.

You can do it even online very comfortably and with nice animation.

Nifty online simulator! My estimates were as much for simplicity as anything (as in _all other things being equal_ (which for Melbourne is Stage 4 lockdown until September 28, and not a large relaxation of restrictions until October 23). This is why I'll be intrigued to see whether the charting of 14 day averages (and yes, it's very easy to fit curves to historical data) will pan out over the coming few weeks. It'll also be interesting to see whether the government officials will adjust their roadmap should these predictions quicken the reduction in cases, but that's an entirely different can of worms.

At the very least it was a fun exercise with an HP Prime, and that's never bad!