DerDangDerDang

Against massive initial skepticism, I’m becoming very impressed with Rust. I’m now going to be ‘that guy’ - apologies.

After years of being annoyed by Rust people claiming C++ is dying or too unsafe or whatever, I expected this to be another C# - a sort of C-with-training-wheels for people who care more about productivity than perf. Something I could safely ignore.

This ‘borrow checker’ stuff sounds great for kids, but for disciplined experienced C++ guys it’s at best a non feature and at worst would just get in the way, like GC, right?

In fact, I’m finding the reduced mental overhead (not just from borrow, but from the language as a whole being smaller and more consistent) is a huge win. Even learning the language and stdlib as I go.

I’ll still use C++ proudly where it’s the best choice, but I’d put some effort into trying Rust first. As a cpp-master-race guy, that surprised the crap out of me.

Thanks, much appreciated.

That’s the sort of thing I expect - the OS starves the program, not the other way around.

Sure, it just seems odd that the OS lets itself get into a situation where it can't recover - even in a corner case like this.

Like, can I just start 20k threads in WebAssembly and reboot anyone who visits my site?

Obviously I'd expect the browser to guard against that... but I expected the OS to as well, so my expectations may be way off!

You’d expect it to kill the offending user space process rather than the OS though, right?

Yeah, that’s pretty worrying that the OS just punts. I always thought limiting the potential damage a user space process could do was one of the main jobs of an OS.

If you have any more OSes laying around to run the test on, I’d be interested to hear how well windows and Linux handle the same thing.

Because on the face of it this seems like a serious bug in the OS. I’m only used to seeing this sort of thing with bad drivers

Thanks, that's enlightening. I am definitely not a statistician, as you've no doubt guessed!

>I think what you may be misunderstanding, and therefore misrepresenting, is the idea that a combination of statistically sampled tests (ONS) plus biased targeted tests (NHS, key workers, Test & Trace etc) results in "skewed" or more misleading R estimates than just the statistically sampled tests (ONS) by themselves.

Not quite, and I apologise if I've communicated poorly. Totally agree that multiple sampling methods lead to more convincing results. I avoid optimising until I've profiled with both a sampling AND an instrumenting profiler, for example.

What I was trying (failing!) to get across is that I think that if the distribution of 'generally available' tests is sufficiently non-representative, then the statistically sampled test model potentially couldn't correct for it enough for the derived estimate to be reliable.

Obviously the responsible subgroup at SAGE will have tried to weight accordingly, and the people running the individual studies will as you say, be much more skilled at accounting for potential bias than I am. I'm absolutely not saying that I've stumbled on some insight that they missed!

Here's where I struggle - the pool of generally available tests at the time fall into two relatively narrow groups: patients exhibiting symptoms bad enough to require hospitalisation, and people at high risk of transmission. The random longitudinal studies don't have that problem of course and should be used to correct for the bias. I realise there are methods for quantifying bias and uncertainty, but isn't this more at risk of inaccuracy than simply having a higher N? N matters, surely? The higher confidence we have that any one set of data is representative, the higher confidence we can have in the final derived estimate.

I just can't see any angle from which having tested more people we wouldn't have improved our ability to estimate R.

That was great, thank you.

I know consciously expressing gratitude can be a powerful way to ground yourself and stay positive, but your insight about spurring action to fight thermodynamics is awesome.

Thanks, I appreciate it.

It does say they've been doing random sampling. They also say that randomly sampled longitudinal surveys are the best predictors. They also list (p32) all the other data sources that are used to estimate R in different studies - including case numbers from tests taken by symptomatic members of the general public.

Section 7.1 Case Numbers notes that Pillar 1 tests aren't a good predictor, Pillar 2 and 3 can be useful but (heavily paraphrasing here if you'll excuse me) the results need to be adjusted and hedged.

This page says they take many studies as input to the official R, including studies that look at test figures in the general population - https://www.gov.uk/guidance/the-r-number-in-the-uk

Section 8 of the rs paper specifically says that longitudinal studies should be used on a much wider scale than they are being, because of asymptomatic transmission. So in August, there was not enough random longitudinal testing, and an over-reliance on studies based on case numbers!

My controversial comment said that early on, access to testing would have skewed things, making R harder to confidently estimate (better wording this time?). I.E. Pillar 2 and 3 tests were not available to the non-essential members of the general public early on.

Is it fair to say that makomk's comment "one of the main things used to estimate the value of R is randomly sampling ... regardless of symptoms" is broadly true (now) - but that my comments about poor early access to testing skewing early R estimates is also broadly true? They're both supported by the paper you linked, imo.

Edit: To be clear, I was wrong about there being no access to testing without symptoms, as at least some random longitudinal testing was happening early on - just not enough, according to the paper you linked.

Thanks for reading ;)

Ok, I really am arguing in good faith here. Not intending to spread disinformation. Would you mind spending the time to conclusively correct me?

So. Are you saying that the R number given by the government is derived from data only from random sampling, and that other test statistics do not feed into the estimate at all?

I’m not a statistician or epidemiologist and I find that very hard to believe. But I will honestly be happy to have learned something if you can teach me otherwise.

My understanding was that longitudinal studies using random sampling were used to check and correct the numbers derived from the actual test numbers for the real population. That meant the longitudinal studies would have been even more important early on when access to testing was poor, because the sample distribution was obviously skewed wrt the whole population.

Again, I really would appreciate a substantial rebuttal, link or otherwise.

Thanks

or how the longitudinal studies numbers are weighted against the overall numbers when coming up with the announced figure..

While you’re right in absolute terms, I think you’re obscuring a valid point.

The ONS having tests doesn’t change the fact that for the vast majority of the population, being an essential worker was the only way to get a test.

The ONS tests are not the only input to the announced R number are they? Maybe I’m misunderstanding horribly, but it seems like you’re saying there’s absolutely no correlation whatsoever between availability of tests to the general public and the accuracy of the announced R number for the general public.

If I’m misunderstanding I do genuinely want to understand!

Absolutely. Not saying they chose the right magic number!

I’m no statistician, but for me the most egregious failures on this have been constant use of ‘The’ R number and ‘The’ curve. The implication is that we have a single infallible measure that’s universally applicable, with no reference to exactly how R from different studies were weighted to come up with the announced figure, or how the sample populations of each study were skewed.

You’re right, and it’s frustrating that this discussion is often framed as a binary choice between lockdown and back to normal.

IMO what we really need to be doing is looking at making lasting changes to society to keep things working safely. We’re seeing a little of that, but not nearly enough.

Exactly that, nominally.

Though pjc50’s comment looks pretty convincing given gov.uk could have used the same money to make it easier for businesses to restructure to offer delivery...

Ah ok. Any idea how extensive the programme is and what weight it's given in relation to overall testing?

Obviously if this is the only / main input to the R number then my previous comment is crap and I retract it. Otherwise, I think I gave reasonable examples of how the sampling has been skewed, in good faith.

Fair point, and you're right something had to give on that score. Classroom pods for young kids seem to work fairly well overall, it just ups the risk for visiting the grandparents.

The university side didn't seem to work out as well - possibly partly because uni staff are (understandably) less willing to work crunch hours to meet duty of care than primary school staff.

It's worth bearing in mind that initially (and until only a few months ago iirc) tests were only available to certified essential workers - first NHS staff, then civil servants, etc. I actually paid £100 for a private test in May. I've had an NHS drive-thru one more recently.

I'm not sure of the absolute veracity of your 'regardless of symptoms' claim. As I understand it, the R number is calculated based on all the NHS tests done (and private, with some lag iirc). This even explicitly says "You cannot get a free NHS test unless you have symptoms, have been asked to by your local council, live in England and have been told to by your hospital, or are taking part in a government pilot project."

https://www.gov.uk/get-coronavirus-test

Wall-to-wall testing would be my preferred route. In fairness I think that would be gov.uk's preferred route too if they could organise it (they can't).

I got temperature check and covid symptom questionnaire on entry to (private) hospital, but it's not widespread.

looks like both of us is one of today's lucky 10,000 ;)

https://dictionary.cambridge.org/dictionary/english/beg-the-...

Headline begs the question - what the hell did we think we were doing between August and November?

Wasn’t R<1 the magic number for ending the first lockdown?

FWIW I’ve always been suspicious of the way R is used by the government and media. It’s always going to be a best guess averaged over an area, but with asymptomatic transmission and initially poor access to testing, seems like it should be treated with caution. Not dismissed, but not the main driver, I guess.

Weird, it’s working ok for me on shitty mobile

"remote control your code"

Looks great, not sure about the pricing. Maybe offer a 50 or 100 dollar package, and charge the full 499 for the ultra fine tuning secret sauce package? I'd understand if you don't want to dilute your content though. Maybe offer as a monthly/yearly subscription? This seems like a topic where regular updates and podcast style discussion could be valuable. Perhaps not to the extent of creating a community, but certainly to the extent of creating a regular audience.

Fair point, and I agree audio is potentially more disruptive - especially if it can get to good real-time performance.

Not sure there’s any deepfakery on the Michael Caine audio. Peter Serafinowicz (the other collaborator and originator of ‘sassy Trump’) is well known for his Michael Caine impersonation.

Good to know, thanks.

Any plans for charts? I found this is the big feature (for my case) supported by Qt but not many others.

Really excited to see more competition in this space!

Flutter looks great, and I seriously considered it for my next project. In the end I decided the possibility of Google getting bored and ending support was enough to tip me to using Qt Quick instead - even though I kinda dislike Qt and js, and will have to pay for it.

I'm not sure it's the responsibility of others to ensure you have enough information not to look silly in a hasty post ;)