Coronavirus thread

[TheRoland]

The code is a disaster. Using the same data, it produces different results with different processors. It produces different results if it runs multithreaded on the same processor. The College seems to think this is a feature because of its pseudo-randomness, yet the randomness was not an intended behavior of the code. This is so clueless as to be dangerous. That's not how software works.

Sure it is. Lack of thread-safety and floating point rounding abound in tons of software, especially simulations produced by scientists. The models that track hurricanes, predict electricity demand, and anticipate internet usage trends all have similar issues. Because they were written in FORTRAN by grad students years ago. There is no Facebook-Amazon-Apple-Netflix-Google team writing any of these things and when there is (like internet trends) it's still data-scientists who write them.

But, again, I know how to write and review this stuff and the reviewer hit my usual checklist points.

To be clear, you're a programmer, but do you often deal with data science and simulation and modeling? I see many models in my day job, and this is absolutely typical and I'm shocked that you're shocked.

[David S.]

I'd be curious as to where you heard that from. The idea of a "zinc ionophore" is new to me and not consistent with what I know about how HCQ work. The concept of an ionophore as antiviral also doesn't really pass the smell test per se... elemental Zinc doesn't really "do" anything on its own in living systems, but it does permit endogenous enzymes to function normally when it is present in physiologic concentrations. Your cells also have very, very complex and responsive machinery to ensure that the overall concentration of dissolved ions (including zinc) are within a very specific range that is compatible with life. Just doesn't come together to form a coherent picture to me.

Here's a March 10 video, so a bit dated now, that seems to explain the theory in layman-ish terms. I'm not educated enough to determine the accuracy of the info.

https://youtu.be/U7F1cnWup9M?t=90

[MickAK]

Not a programmer or an epidemiologist, but generally speaking academics are researchers first and programmers (and writers, and public speakers, and )... 2nd. I can definitely confirm there are not millions of dollars to throw around for a project like this. Presumably it was put together by someone who is an epidemiologist first and a programmer second, so it's not a surprise they aren't silicon valley quality products. They also are likely working with a relatively small team, of whom some are graduate students/post-doctoral fellows/otherwise in training.

That said, they usually have their jobs because they are not capable of imaging themselves doing something more profitable and/or less stressful. This usually means they're pretty good at what they do. I would not take a programmer's evaluation of their model seriously any more than I would take a book critic's evaluation of a scientific article. Sure, it's written in english, but the quality of the prose is very far from the point.

That is not really how code works. The issue is not that they aren't 'Silicon Valley quality products'. The issue is that the models are wrong because the code is bad. There can be laughably bad code that does the job just fine, it's just difficult to work with. That doesn't really matter. If the code errors end up influencing the results in a way that makes the results almost meaningless, then using it as a basis for public policy is a terrible idea.

[Edster]

To be clear, you're a programmer, but do you often deal with data science and simulation and modeling? I see many models in my day job, and this is absolutely typical and I'm shocked that you're shocked.

Yes. I have worked with and overseen programmers developing modeling and forecasting software in private industry. Your point regarding my naivete outside of my setting is probably valid.

In my world, an incorrect forecast and sloppy code have financial consequences that are identified quickly and not long tolerated. To just say it: I would have fired some people. But I would be expected to do that in my situation.

I'll add that the reviewer of the Imperial code suggested this type of software be implemented by people who have written it for insurance companies. Again, this is a world where being unqualified, sloppy, or wrong tends to make you unemployed.

For research and sandbox purposes, quick and dirty is fine. But this isn't like the MySQL database and Python scripts folks set up to track sports stats. Lives were impacted here. The stakes are too high for an unprofessional approach. This is needing to build a four-lane interstate bridge across a river and saying, "Heck with it, let's go buy a few bags of cement and get started."

[Nephrology]

Here's a March 10 video, so a bit dated now, that seems to explain the theory in layman-ish terms. I'm not educated enough to determine the accuracy of the info.

https://youtu.be/U7F1cnWup9M?t=90

Huh, that's interesting. This article they link in particular has my attention, but admittedly only because I respect the journal it is published in. I have not had the chance to read and weigh their arguments on their own merit yet. I'll take a close look at it later - have some sort of School of Medicine Zoom call coming up shortly.

[TheRoland]

I don’t really think we’re disagreeing with each other, but if you want a different modeling approach you need a time machine and millions in grant money that nobody could justify spending a year ago.

[bravo7]

https://youtu.be/t8kzTSVTxVQ

12 min interview from a doctor with a reasonable approach to the current state of the virus.

[0ddl0t]

Apparently after today's uproar this has since been amended so only those who were hospitalized are permanently disqualified, but still makes you wonder what they know...

[Nephrology]

Apparently after today's uproar this has since been amended so only those who were hospitalized are permanently disqualified, but still makes you wonder what they know...

What are you implying exactly? I don't follow.

[Caballoflaco]

What are you implying exactly? I don't follow.

In a preliminary draft of recruiting guidelines they made any history of covid-19 an automatic disqualifier for military service. At first glance I was like wtf too, why would you do that?