RSS
As I like to point out, progressive broadsheets like the New York Times and the Guardian go out of their way to make the news less interesting. Here’s the Guardian on a hilarious scandal:
UK scientists being drawn into ‘very unpleasant’ political situation
Colleagues raise concerns after Prof Neil Ferguson stepped down from advisory role
Neil Ferguson is an epidemiologist at Imperial College London who was a top adviser to the UK government. He is, by the way, not the same person as financial historian Niall Ferguson.
Ian Sample, Kate Proctor and Rowena Mason
Wed 6 May 2020
Ferguson had become one of the most trusted voices on the outbreak.
Scientists who advise the government on its coronavirus strategy have warned they are being drawn into politics after the leading infectious disease modeller Prof Neil Ferguson stepped down as one of the cabinet’s most prominent advisers.
Ferguson, the head of the Imperial College team whose modelling persuaded ministers that Britain needed to order a lockdown to prevent hundreds of thousands of deaths, resigned from the Scientific Advisory Group for Emergencies (Sage) on Tuesday night.
Ferguson left the position after it emerged that his lover had visited his home on two occasions in March and April in a breach of the government’s official guidance on social distancing.
One scientific adviser to the government said Ferguson’s resignation had created “an awful lot of concern” and that the mood in the community was “very depressed”. The events revealed how university academics who lent their advice to government were having to cope with an increasingly difficult situation, the adviser added.
“He’s an academic researcher. He doesn’t make decisions. He’s not paid for any of this. We are being drawn into a political situation which is very unpleasant,” they said.
The prurient headlines about Neil Ferguson are a huge distraction
Ferguson had become one of the most trusted voices on the outbreak and took a prime role in discussing the reasons for the lockdown and social distancing measures. But that prominence led to him being portrayed as a decision-maker rather than an independent scientist whose team generated predictions that informed ministers about what strategies they might take.
“The media are presenting the scientists as decision-makers when they’re not. We are in the frontline when we shouldn’t be. We are not the story,” the adviser said. Unlike the government’s chief scientific adviser, Sir Patrick Vallance, and the chief medical officer, Professor Chris Whitty, the independent university academics who sit on Sage do not have teams to manage their media appearances and support them.
Scientists put forward a conspiracy theory:
Prof Anthony Costello, a former director at the World Health Organization, questioned why the story about Ferguson breaching social distancing rules a month ago was reported in the Daily Telegraph on the day that Britain’s official death rate overtook Italy’s, and before an imminent decision to relax the lockdown. Another scientific adviser who spoke to the Guardian said: “Ferguson was on the front pages when Britain having the highest death rate in Europe should have been splashed everywhere.”
The Telegraph revealed that Antonia Staats had crossed London from her family home to visit Ferguson on at least two occasions, on 30 March and 8 April, since lockdown measures were imposed.
Neil Ferguson….. take a look at his wife. Whoa! I did a double take. No wonder this poor sap strayed. I will give him credit for doing the manly thing during tuff times. Don’t cry for this jamoke, he is still gettin’ paid from his university professor job.
Time Magazine has the cover story all ready to go . . .
https://api.time.com/wp-content/uploads/2014/08/ferguson-cover.jpg?w=800&quality=85
Staats said she thought that they were one “household.” Certainly Ferguson knows better than this? But I think it’s true that the concept of household quarantine is not understood by many civilians. They seem shocked to learn that, unless they live under that same roof, one’s household
does not include family, friends, lovers, housekeepers, handymen, babysitters, or children’s playmates.
This should be explained more clearly by the authorities, and the underlying reasons should be explained, perhaps with some exponential math and modeling examples.
Who? Whom? Scum of the earth.
Are these real people?
I’m as optimistic about Steve’s return to Noticing as I am about having a draft beer. Hopefully within days reach.
Ron Unz, ladies and gentlemen. The man who takes pride in his ability to sniff out propaganda and lies.
I’m thinking of a movie script. A group of idiots who believe the NYSLIMES and the rest of the media bug out to some prepper hideout. The hideout was well organized in advance for some end of the world event such as the 2019/20 flu season.
For 30 years the idiots grow their own food, get around on bikes and run their happy little self sufficient society. The children grew up believing the world outside is a vast wasteland of infected rotten corpses. About page 30 there’s a precipating event. The young adults have to go out into the diseased world. They discover there was no plague, few died, average age of the dead 79 and life on earth is better than it was when their idiot parents took them off to the hideout,Replies: @AnotherDad, @jsm, @James N. Kennett
Ferguson has a terrible track record going back at least 20 years. So naturally he’s the government’s go to guy on this:
I've worked in casinos and restaurants for most of the last decade or so....with as much as they ride my ass if I had performed like that I wouldn't have a job now...
There is obvioulsy a great market for credentialed alarmists. Even more so in the UK, the ultimate nanny state, than in the U.S.
I’m not sure i follow what inversion you’re pointing to here.
The left MSM is still strongly on board with the “panic” side. So they have to defend the lockdown. The original story about Ferguson showed the hypocrisy of one of their lockdown saints so they have to set deflectors to full while they de-canonize him.
“Neil Ferguson is an epidemiologist at Imperial College London who was a top adviser to the UK government. He is, by the way, not the same person as financial historian Niall Ferguson.”
I keep confusing the two, and it would be hilarious if they were the same person. However, the historian Ferguson, while also a courtier, is a fairly competent historian.
Here is an article on the code for his inaccurate model: https://lockdownsceptics.org/code-review-of-fergusons-model/
I don’t know anything about coding, but it doesn’t sound great for Mr. Fuergeson.
TBH what he describes is not as uncommon as one would think in software "out there", including in the financial system.
Academic environments are notorious for this kind of code because their focus and expertise is not "software engineering", so you've probably had a countless grad students and interns messing with these things for years with nobody enforcing standards or uniformity. And each one is primarily focused on implementing/changing whatever their thesis is about, not on doing house-keeping, documentation, testing, etc which is generally thankless even in commercial shops. Writing maintainable code / projects just isn't on most academics' radar.Replies: @Kratoklastes
But if you're going to simulate, your assumptions should be part of any consideration of your output. And reusing old code is definitely a risk.
The first Ariane 5 rockets blew itself up because its Ariane 4 software didn't like to accelerate so fast.Replies: @Polynikes, @JosephB
The quality of the code will be hugely embarrassing for Imperial College, and would not have been acceptable in the government departments that Ferguson was advising - they adopted coding standards decades ago.
Whether the output of the code is trustworthy is another matter.
I took a snapshot of the ZeroHedge Headline of this story:
“Scientist Whose Doomsday Models Sparked Global Lockdown Resigns After Breaking Quarantine to Bang Married Lover”.
Which goes to show you, Marx was right – history repeats, first as tragedy, then as farce.
The interesting thing is that I have shown that headline to eight men, all of whom laughed out loud, and three women, all of whom viciously attacked me personally. Make of that what you will. The thing I make of it is this: The women driven to rage against me are not only defending this guys execrable double standard, but his involvement in adultery as well.
Ferguson may be a public figure but his girlfriend is not, and so didn’t deserve to have her picture and name splashed all over the news media. Her husband and children, even less so.
Ferguson himself might have been thinking along the lines of Dr Strangelove where he suggests that key officials be supplied with multiple fertile women to help repopulate the human race.
She is an "organizer" for a Soros-funded activist group.Replies: @Stan d Mute
Ferguson himself might have been thinking along the lines of Dr Strangelove where he suggests that key officials be supplied with multiple fertile women to help repopulate the human race.Replies: @O'Really, @Anonymous, @Jenner Ickham Errican, @Mr. Anon, @Alexander Turok, @Bardon Kaldian
The girlfriend actually is a public figure.
She is an “organizer” for a Soros-funded activist group.
Doesn’t Soros have a history of profit from destruction he’s engineered?
Ferguson himself might have been thinking along the lines of Dr Strangelove where he suggests that key officials be supplied with multiple fertile women to help repopulate the human race.Replies: @O'Really, @Anonymous, @Jenner Ickham Errican, @Mr. Anon, @Alexander Turok, @Bardon Kaldian
Apparently the woman is some sort of hardcore leftist, Greta type climate activist. Naturally, that kind of woman is going to be impressed by government lockdowns that impose economic austerity and by men like Ferguson who can have power and influence over such government actions. I wonder if that played a role in Ferguson’s thinking and behavior, and if it induced him if only a little bit more to recommend a lockdown in order to impress her.
"masters in Asian politics from the School of Oriental and African Studies"
ALL THE STAATS: Who is Antonia Staats and who’s her husband?
https://www.thesun.co.uk/news/11559162/who-is-antonia-staats-husband/
Ferguson himself might have been thinking along the lines of Dr Strangelove where he suggests that key officials be supplied with multiple fertile women to help repopulate the human race.Replies: @O'Really, @Anonymous, @Jenner Ickham Errican, @Mr. Anon, @Alexander Turok, @Bardon Kaldian
LOL. If one eschews the public eyeball, one avoids public figure frucking.
Has someone checked the headlines in The Sun, The Star, The Mirror, The Daily Mail, The Daily Express, and any other British papers.
The Financial Times story might be snooze worthy (or entertainingly subtle?).
I’m off to the Spectator UK.
"Scientist Whose Doomsday Models Sparked Global Lockdown Resigns After Breaking Quarantine to Bang Married Lover".
Which goes to show you, Marx was right - history repeats, first as tragedy, then as farce.
The interesting thing is that I have shown that headline to eight men, all of whom laughed out loud, and three women, all of whom viciously attacked me personally. Make of that what you will. The thing I make of it is this: The women driven to rage against me are not only defending this guys execrable double standard, but his involvement in adultery as well.Replies: @Kylie, @vhrm, @J.Ross
Why did the women attack you for showing them this headline?
How wrong was Imperial College forecast? March 16 document estimated daily deaths at about 20/100,000 at the peak in the UK which would amount to 13,200/day at the peak. This forecast was for the unconstrained outbreak.
https://www.imperial.ac.uk/media/imperial-college/medicine/sph/ide/gida-fellowships/Imperial-College-COVID19-NPI-modelling-16-03-2020.pdf
Providing that Sweden indeed is an example of an unconstrained outbreak (which is not entirely true) and comparing it with Finland and Norway, two countries of similar population densities that implemented the countermeasures, we can estimate the factor by which the countermeasure reduce the daily deaths. This factor is 7.5 for Sweden and Scandinavian countries (population density corrected).
If we apply 7.5 to 13,200 we get 1,760 deaths/day at the peak. Now we look at UK daily deaths at worldometers.info and we see that at maximum numbers are around 1,100 per day. It is better than factor of 2. I would say that Imperial College forecast was pretty good to be off only by 60% and I do not understand why so many people are so upset with it. It did its job. It got the UK to implement the countermeasures that saved so far 29,000*(7.5-1)=188,500 lives in the UK.
Ferguson himself might have been thinking along the lines of Dr Strangelove where he suggests that key officials be supplied with multiple fertile women to help repopulate the human race.Replies: @O'Really, @Anonymous, @Jenner Ickham Errican, @Mr. Anon, @Alexander Turok, @Bardon Kaldian
She’s a professional “climate activist”. She’s a player.
Is the inversion that he said he had recovered from covid 19 and was not infectious? That was reported in some accounts.
I don’t know anything about coding, but it doesn’t sound great for Mr. Fuergeson.Replies: @vhrm, @David, @Jim Don Bob, @James N. Kennett
excellent, thanks for posting this.
TBH what he describes is not as uncommon as one would think in software “out there”, including in the financial system.
Academic environments are notorious for this kind of code because their focus and expertise is not “software engineering”, so you’ve probably had a countless grad students and interns messing with these things for years with nobody enforcing standards or uniformity. And each one is primarily focused on implementing/changing whatever their thesis is about, not on doing house-keeping, documentation, testing, etc which is generally thankless even in commercial shops. Writing maintainable code / projects just isn’t on most academics’ radar.
When your team claims to be specialised in a specific type of modelling, and has a 'base' model that is claimed to be state of the art, then that motherfucker had better be able to be picked up by an numerate layman and understood directly from the documentation.
I would go further: such a 'shop' ought to have a 'toy' version of their model which gives the aforementioned numerate layman a way to understand the stylised facts of the 'big' model.
That's been the 'house view' at CoPS since the 1980s. I was there in the 1990s as an RA and PhD student.
As a very good example: go to the webpage for the very large number of variants of ORANI-G; download any of the ZIP files, and examine any of the TAB files and look how structured it is.
Then look at the DOC file in the package and you will get a full explanation of each of the models equations and variables.
COMPLETE documentation. NO code-spaghetti.
That is the standard expected from a professional group of academic modellers with a 40-year history of global leadership in their discipline. It is at it was when I joined; it was the same when I left in 2000, and it's the same today.
That said: the ICL model code isn't that badly structured. I've seen much much worse - in quant finance there is a very large amount of badly-structured, badly-documented code, that sometimes doesn't even do what it's supposed to (it passes rudimentary tests by accident as a result of some corner case, and then is viewed as 'correct' and then goes into production).
Take a look at this single equation from a MATLAB model that purported to calculate the sector-level returns on a portfolio... note that the red apostrophe indicates a transpose, and the text in blue is effectively a subscript (it's the range - from 1 to the maximum months of data available for the first benchmark for the fund). Originally there were no spaces in the line -
https://www.dropbox.com/s/anbogp8ah3psxwj/ARBS.png?dl=1
Turns out that, both of the quantities on the RHS were already calculated elsewhere (again, on the fly) and were never stored. This happened no less than 8 times in the code, sometimes using different names for the same variables.
So after refactoring, and declaring a convenience variable, the above shitshow becomesAnyhow... refactoring bad code is also a terrific way to work out what it does - and compare that to what the code's owners think it should do.
.
My disagreement with the ICL model is not its stylistic shortcomings; it's the lack of sensitivity analysis that was performed before Ferguson started telling the political class a horror story that jibed with their desire to be seen to save the world. It's the same shit he has pulled roughly twice a decade since the mid-90s - which indicates it wasn't an accident.Replies: @jbwilson24
"Scientist Whose Doomsday Models Sparked Global Lockdown Resigns After Breaking Quarantine to Bang Married Lover".
Which goes to show you, Marx was right - history repeats, first as tragedy, then as farce.
The interesting thing is that I have shown that headline to eight men, all of whom laughed out loud, and three women, all of whom viciously attacked me personally. Make of that what you will. The thing I make of it is this: The women driven to rage against me are not only defending this guys execrable double standard, but his involvement in adultery as well.Replies: @Kylie, @vhrm, @J.Ross
Note that the woman is supposedly in an open marriage. That’s risque in it’s own way, but it’s not the standard narrative.
https://i.pinimg.com/originals/e9/c8/f2/e9c8f2181a21f15bf6d4b9a914276445.gif
Antonia Staats Husband (The Cuckold)
Antonia is in an open marriage with Chris Lucas. In late March she said in a podcast interview that the lockdown presented an “interesting relationship challenge”.
“I think it’s also a strain on – maybe strained has sounded too negative – but it’s an interesting relationship challenge, for [her husband] and my relationship.”
Neil Ferguson Mistress (Does she count as a mistress if Ferguson is not married?)
Staats is a lover of Neil Ferguson who is a British epidemiologist and professor of mathematical biology at Imperial College London. Neil served as the advisor of Boris Johnson, Prime Minister of The UK and was a prominent member of the Scientific Advisory Group for Emergencies (SAGE) which has been spearheading the country’s coronavirus response. He convinced Boris Johnson to order for a lockdown but he defied the rule by asking his mistress to visit him twice. He had appeared before the Commons Science and Technology Committee on March 25 warning that NHS will be overwhelmed without a lockdown.
On March 30, Antonia visited him at his house, after finishing two of self isolation as he had tested positive for coronavirus.
On April 4, he spoke to BBC Radio 4’s Today where he said; ‘We want to move to a situation where at least by the end of May we can substitute less intensive measures for the current lockdown we have now… I don’t think anyone wants to lift measures at the current time and risk the epidemic getting worse’
On April 8, Antonia again visited Ferguson despite telling friends that her husband was showing symptoms of coronavirus.
Neil Ferguson Resigns
Piers Morgan tweeted: ‘Unbelievable and shocking. So, the government is ”following the science” of scientists who don’t even follow their own science. What a shameful shambles. Professor Ferguson’s excuse is he thought he was immune from COVID-19 after having it – despite there being zero scientific proof people who’ve had it actually get immunity. And this guy’s the No1 ‘expert’ on whom the government is basing its entire coronavirus strategy?’
They could be from a Kingsley Amis novel…
Here’s another theory: Ferguson is a serial hysteric and alarmist (a good man to have in a crisis – not!). He calculated that the end of the world was nigh, so he decided, “Why not go out with an adulterous bang rather than a whimper?”
In his position, if you believed his pants-wetting prediction, wouldn’t you do the same?
Had the exponent predictions and modeling of 2 months ago on this site come true, there’d be no more site as every author and commenter would be dead and there’d be no internet because every tech worker on the planet would be dead.
I’m thinking of a movie script. A group of idiots who believe the NYSLIMES and the rest of the media bug out to some prepper hideout. The hideout was well organized in advance for some end of the world event such as the 2019/20 flu season.
For 30 years the idiots grow their own food, get around on bikes and run their happy little self sufficient society. The children grew up believing the world outside is a vast wasteland of infected rotten corpses. About page 30 there’s a precipating event. The young adults have to go out into the diseased world. They discover there was no plague, few died, average age of the dead 79 and life on earth is better than it was when their idiot parents took them off to the hideout,
I thought those predictions were over-the-top because the Diamond Princess data established that there were a lot of asymptomatic infections. So the real death rate would be lower than what was reported.
But if that was not true (and you had a full 1% IFR) and this thing exploded--no reaction, no social distancing, no lockdown, no masks--then a couple million dead before getting to herd immunity wasn't ridiculous at all.
But that doesn't mean it was ever going to be an apocalypse. 2.8 million Americans--mostly very old--die every year. This thing killing another 2 or even 3 or 4 million--somewhat younger, but still mostly very old Americans--also would not have been the apocalypse, nor brought on any of the things you mention.
~~
I like your movie script.
However, the real kicker would be that when these kids re-discover the world ... they'll at first think their parents were idiots and run out into the big "modern" world ....
but then find out their little world of hard work is actually healthier:
No fats and tats, better behavior, more love, healthier marriages, higher fertility, more children, a brighter future.
The payoff will be most of them deciding to go back into hideout and resume healthy living.Replies: @Clyde, @TomSchmidt
Blast from the Past starring Brendan Fraser. Just, the apocalypse was nuclear war that didn't happen
Hahahahhahahhahahhaaahaaaa Boring was my word! Saying something is ‘boring’ ends conversations/squabbles with families/destroys Democrats! It is used by someone feeling sophomoric or pugilistic, like me….a lot, these days. I just don’t want the USA to turn into a fascist state, a technocrat fascist state.
Sorry I’m late!
The Guardian put quotes around the remark, but aren’t telling us who (if anyone) might actually have uttered them.
Um, who said this? Another Easter Bunny?
No quotes, so at least the Guardian is admitting they said this themselves.
Just like that Tara Reade, who’s trying to distract This Entire Nation from the Glorious Goal of installing What’s His Name to take the place of Bad Orange Man.
Ran out of edit window.
It also need pointing out that it’s “two occasions” they admit to so far and “his lover” omits the key word “married” because I’m just old-fashioned like that.
"Scientist Whose Doomsday Models Sparked Global Lockdown Resigns After Breaking Quarantine to Bang Married Lover".
Which goes to show you, Marx was right - history repeats, first as tragedy, then as farce.
The interesting thing is that I have shown that headline to eight men, all of whom laughed out loud, and three women, all of whom viciously attacked me personally. Make of that what you will. The thing I make of it is this: The women driven to rage against me are not only defending this guys execrable double standard, but his involvement in adultery as well.Replies: @Kylie, @vhrm, @J.Ross
Back when … uhhh … Football Hero, who is married to Gisele Bundchen, cheated on her (the Patriots way: gratuitous cheating when you’re winning anyway), Chateau Heartiste linked evidence that women were totally okay with it. Some furthermore said that if they were in Gisele’s position they would stay with [that guy].
The good professor should have just explained:
“I just wanted to do my part to flatten her curves.”
The New York Times and the Guardian may indeed try and make this sort of news less interesting, but even so, I can’t see the New York Times using the blunt and direct term “lover”, in its reportage, as the British Guardian newspaper does.
The Daily Mail, of course, has a lot more fun than the Guardian, uses the even more direct term “mistress”, and has the usual rich complement of photographs.
https://www.dailymail.co.uk/news/article-8294315/Naked-hypocrisy-Professor-Lockdowns-mistress-telling-public-stay-home.html
PS: Curiously, Neil Ferguson strongly resembles the woman’s husband.
She is an "organizer" for a Soros-funded activist group.Replies: @Stan d Mute
Ah, and now we potentially have an answer to the question, “Cui bono?”
Doesn’t Soros have a history of profit from destruction he’s engineered?
The left MSM is still strongly on board with the "panic" side. So they have to defend the lockdown. The original story about Ferguson showed the hypocrisy of one of their lockdown saints so they have to set deflectors to full while they de-canonize him.Replies: @Federalist
The CoronaPanic narrative is unraveling. In retrospect, Ferguson’s wild predictions are now particularly problematic. Rather than letting Ferguson discredit the narrative, they have decided to discredit him (or de-canonize him as you said) on other grounds. Now when his models or predictions are attacked, they don’t have to defend him. He’s already been discredited as a quarantine-breaking adulterer. The story is now about how Ferguson was screwing some woman when he was supposed to be isolated. The story is not about why the hell did our rulers cause an economic catastrophe listening to a guy who has been wrong about everything.
The Financial Times story might be snooze worthy (or entertainingly subtle?).
I’m off to the Spectator UK.Replies: @Cowboy Shaw
The front page headline in The Sun was: “Prof Lockdown broke lockdown to get his trousers down.”
Ferguson himself might have been thinking along the lines of Dr Strangelove where he suggests that key officials be supplied with multiple fertile women to help repopulate the human race.Replies: @O'Really, @Anonymous, @Jenner Ickham Errican, @Mr. Anon, @Alexander Turok, @Bardon Kaldian
This sympathy for the privacy of adulterers has always puzzled me. If she’s feeling a lot of shame, well, she should be feeling ashamed. She’s doing something immoral. But people insist on seeing it as a victimless crime.
Perhaps we could split the institution of marriage in two: one for people who think the commitment means something and another for those who don’t.
there is no tar and feathering, then for miscreants, who are then driven out of town on a mule anymore is there? 21st century man is so enlightened.
oh wait silly me it has happened …
https://www.hoover.org/research/politically-incorrect-guide-science
These conspiracy theorists should not be afforded a public platform … he is probably given to dressing up in private in Hugo Boss gear while listening to his Hitler’s Greatest Orations CD.
Mark Zuckerberg’s newly convened Star Chamber should look into his credentials, pronto, with a view to getting him de-personed:
https://www.dailymail.co.uk/news/article-8293777/Facebook-reveals-members-supreme-court-content.html
As real as Neil “Booty Call” Ferguson, his nickname when he gets back to the lab.
This guy Ferguson should be an incel.
The company his lover Antonia Staat works for – Avaaz – seem pretty slimy. Some kind of progressive cult. The founder is Ricken Patel, an Indian-Jewish Canadian. The reviews on Glassdoor for Avaaz say it all:
I’m sure a bit of digging will discover that Avvaaz is a NXIVM-style sex cult which attracts ‘open marriage’ advocates.
What do you get if you put a historian, a cryptographer, and an epidemiologist into a blender?
So an historian, a cryptographer, and an epidemiologist go duck hunting.
A duck rises from the lake. The cryptographer jerks his gun up and fires 10 feet in front of the duck. "I was sure that's where it was going" he murmurs, disappointed.
The historian takes slow, careful aim, tracking the duck over a quarter of the sky, fires and misses ten feet behind the duck. "I'm quite positive that's where it was," he says dejectedly.
The epidemiologist watches the duck fly off and says quietly
"I'm sure we hit it once."
That could be quite the remake: The Man Who Vaccinated “Liberty” Vallance
Say, didn’t “wife-swapping” originate with WWII pilots?
How ‘Professor Lockdown’ fell for German lover who is married to a Battle of Britain pilot’s grandson: Neil Ferguson had secret trysts with left-wing campaigner in an ‘open marriage’ with fellow academic
Her surname is a palindrome. How often do you see that? Do they have a child named Sacul? Sacul Staats Lucas.
Antonia Staats = Saint to a Satan.
I don’t know anything about coding, but it doesn’t sound great for Mr. Fuergeson.Replies: @vhrm, @David, @Jim Don Bob, @James N. Kennett
I think that guy is being a little silly about simulation. Like implying that there is something absurd in using the word stochastic vs random. Or acting surprised that a simulation model doesn’t produce the same result every time, and that one’s final “answer” might be an average of several runs.
But if you’re going to simulate, your assumptions should be part of any consideration of your output. And reusing old code is definitely a risk.
The first Ariane 5 rockets blew itself up because its Ariane 4 software didn’t like to accelerate so fast.
For example, the IHME model could get hard data on day to day hospital admissions which was part of their chart. They would release updates that were off by as much as 50% for their next day forecasts. That is almost impossible. A child drawing a straight line on the graph could've had more accurate guesses. Similarly, Wisconsin released one of their projections that supposedly justifies the shutdown; it had projections for dates already passed that were up to 100% off. In other words, they couldn't separate themselves from their theory to even reconcile with the reality in front of their face.That article is mostly foreign to me, because I know nothing about coding. Do you get the impression Furgeson was sloppy, stupid, malicious, or some combination thereof?Replies: @David
OMG, the imagery you painted with this is horrifying, and that’s not even counting the terrifying possibility that Thunberg could actually reproduce at some point.
TBH what he describes is not as uncommon as one would think in software "out there", including in the financial system.
Academic environments are notorious for this kind of code because their focus and expertise is not "software engineering", so you've probably had a countless grad students and interns messing with these things for years with nobody enforcing standards or uniformity. And each one is primarily focused on implementing/changing whatever their thesis is about, not on doing house-keeping, documentation, testing, etc which is generally thankless even in commercial shops. Writing maintainable code / projects just isn't on most academics' radar.Replies: @Kratoklastes
None of this is true when your entire team is specifically a numerical modelling team.
When your team claims to be specialised in a specific type of modelling, and has a ‘base’ model that is claimed to be state of the art, then that motherfucker had better be able to be picked up by an numerate layman and understood directly from the documentation.
I would go further: such a ‘shop’ ought to have a ‘toy’ version of their model which gives the aforementioned numerate layman a way to understand the stylised facts of the ‘big’ model.
That’s been the ‘house view’ at CoPS since the 1980s. I was there in the 1990s as an RA and PhD student.
As a very good example: go to the webpage for the very large number of variants of ORANI-G; download any of the ZIP files, and examine any of the TAB files and look how structured it is.
Then look at the DOC file in the package and you will get a full explanation of each of the models equations and variables.
COMPLETE documentation. NO code-spaghetti.
That is the standard expected from a professional group of academic modellers with a 40-year history of global leadership in their discipline. It is at it was when I joined; it was the same when I left in 2000, and it’s the same today.
That said: the ICL model code isn’t that badly structured. I’ve seen much much worse – in quant finance there is a very large amount of badly-structured, badly-documented code, that sometimes doesn’t even do what it’s supposed to (it passes rudimentary tests by accident as a result of some corner case, and then is viewed as ‘correct’ and then goes into production).
Take a look at this single equation from a MATLAB model that purported to calculate the sector-level returns on a portfolio… note that the red apostrophe indicates a transpose, and the text in blue is effectively a subscript (it’s the range – from 1 to the maximum months of data available for the first benchmark for the fund). Originally there were no spaces in the line –
Turns out that, both of the quantities on the RHS were already calculated elsewhere (again, on the fly) and were never stored. This happened no less than 8 times in the code, sometimes using different names for the same variables.
So after refactoring, and declaring a convenience variable, the above shitshow becomes
Anyhow… refactoring bad code is also a terrific way to work out what it does – and compare that to what the code’s owners think it should do.
.
My disagreement with the ICL model is not its stylistic shortcomings; it’s the lack of sensitivity analysis that was performed before Ferguson started telling the political class a horror story that jibed with their desire to be seen to save the world. It’s the same shit he has pulled roughly twice a decade since the mid-90s – which indicates it wasn’t an accident.
There's bad code, then there is complete and utter professional negligence. This model was relied upon for actual policy decisions that impact lives, making it safety-critical.
None of the usual software engineering techniques (circa 2000) appear to have been followed. Unit tests, proper modularization, separation of concerns, etc. He wrote this model in 2007, when Java, XUnit, execution tracers and formal verification were around. There's absolutely no excuse for this complete and utter quackery.
Amateur programmers shouldn't be anywhere near C or any other language that allows direct access to memory. Ever. Java was widely supported back in 2007, and there were numerous simulation toolkits at an appropriate level of abstraction.
Safety-critical software should be treated with extra care. If you look at other forms of modeling, like transport or nuclear, the various modeling groups typically have actual software engineers on staff. That is, people whose entire skillset involves software. Given that the guy received 130 million or so, he surely could have spent some cash on a couple of actual software engineers.
Unless you have worked for a proper software development company where you'd get yelled at (or poorly reviewed) for bad software technique, you are an amateur. Software is a craft that requires discipline, and you learn by being immersed in an environment where you are guided/smacked by more experienced developers. Amateurs need to take extra care not to do stupid things, but this guy seems to have reveled in his incompetence.
(I won't even mention other modeling issues, such as various forms of uncertainty and model error that are widely discussed in proper fields like climate or fluid modeling).
He should face discipline for incompetence. The episode with his lover is of little consequence in comparison.Replies: @vhrm, @Kratoklastes
Maybe the most curious thing about this whole affair regarding Neil Ferguson, is the way the media picked a rather ‘glamour shot’ of his girlfriend Antonia Staats to highlight
Here’s her more typical appearance in daily life
Imagine if the blonde bombshell had a Russian surname, folk would immediately jump to the conclusion that she was working for Putin. Instead we find she works for a globalist pressure group pushing internet censorship and the climate crisis line. So it looks like poor Ferguson fell – willingly no doubt – into a classic honeytrap. Lucky guy, unlucky us.
Within liberalism, science ultimately serve as theology. An attack of the individual authority of a scientist is perceived as an attack on the majesty of the priesthood.
Oh great. An obscure leftist wacko single-handedly ruins an entire nation (and lends a hand in the ruin abroad) because she has a pretty face.
“masters in Asian politics from the School of Oriental and African Studies”
ALL THE STAATS: Who is Antonia Staats and who’s her husband?
https://www.thesun.co.uk/news/11559162/who-is-antonia-staats-husband/
An historian, a cryptographer, and an epidemiologist walk into a bar….
Open marriage? I don’t get why they still call it a marriage ….
But if you're going to simulate, your assumptions should be part of any consideration of your output. And reusing old code is definitely a risk.
The first Ariane 5 rockets blew itself up because its Ariane 4 software didn't like to accelerate so fast.Replies: @Polynikes, @JosephB
This is true of any simulation or statistical model and where it seems most of our “experts” went wrong. The “super forecasters” talk about rigorously testing their assumptions and re-testing them. It seems with most of these faulty models they took an initial set of numbers from China, plugged them in and went wild with whatever crazy nonsense was spit out the other side.
For example, the IHME model could get hard data on day to day hospital admissions which was part of their chart. They would release updates that were off by as much as 50% for their next day forecasts. That is almost impossible. A child drawing a straight line on the graph could’ve had more accurate guesses. Similarly, Wisconsin released one of their projections that supposedly justifies the shutdown; it had projections for dates already passed that were up to 100% off. In other words, they couldn’t separate themselves from their theory to even reconcile with the reality in front of their face.
That article is mostly foreign to me, because I know nothing about coding. Do you get the impression Furgeson was sloppy, stupid, malicious, or some combination thereof?
In a college finance text book, I read that if a year into your business plan, you're at half your original revenue target, you should halve all your projected revenues. I've never seen a company do that.
https://fact-files.com/antonia-staats-wiki/
Antonia Staats Husband (The Cuckold)
Antonia is in an open marriage with Chris Lucas. In late March she said in a podcast interview that the lockdown presented an “interesting relationship challenge”.
“I think it’s also a strain on – maybe strained has sounded too negative – but it’s an interesting relationship challenge, for [her husband] and my relationship.”
Neil Ferguson Mistress (Does she count as a mistress if Ferguson is not married?)
Staats is a lover of Neil Ferguson who is a British epidemiologist and professor of mathematical biology at Imperial College London. Neil served as the advisor of Boris Johnson, Prime Minister of The UK and was a prominent member of the Scientific Advisory Group for Emergencies (SAGE) which has been spearheading the country’s coronavirus response. He convinced Boris Johnson to order for a lockdown but he defied the rule by asking his mistress to visit him twice. He had appeared before the Commons Science and Technology Committee on March 25 warning that NHS will be overwhelmed without a lockdown.
On March 30, Antonia visited him at his house, after finishing two of self isolation as he had tested positive for coronavirus.
On April 4, he spoke to BBC Radio 4’s Today where he said; ‘We want to move to a situation where at least by the end of May we can substitute less intensive measures for the current lockdown we have now… I don’t think anyone wants to lift measures at the current time and risk the epidemic getting worse’
On April 8, Antonia again visited Ferguson despite telling friends that her husband was showing symptoms of coronavirus.
Neil Ferguson Resigns
Piers Morgan tweeted: ‘Unbelievable and shocking. So, the government is ”following the science” of scientists who don’t even follow their own science. What a shameful shambles. Professor Ferguson’s excuse is he thought he was immune from COVID-19 after having it – despite there being zero scientific proof people who’ve had it actually get immunity. And this guy’s the No1 ‘expert’ on whom the government is basing its entire coronavirus strategy?’
Here's her more typical appearance in daily life
https://i.ibb.co/zfrqHHM/antonia-staats-activist.jpgReplies: @George, @Known Fact, @Kent Nationalist
Interesting mix of let it all hang out sweats coupled with impractical high healed dress shoes.
Need one point out that the wildly hyperbolic Dr. Doom known as Neil Ferguson is no [nameless] Football Hero? But then, Antonia is no Giselle either (see comment #45 above).
I’m thinking of a movie script. A group of idiots who believe the NYSLIMES and the rest of the media bug out to some prepper hideout. The hideout was well organized in advance for some end of the world event such as the 2019/20 flu season.
For 30 years the idiots grow their own food, get around on bikes and run their happy little self sufficient society. The children grew up believing the world outside is a vast wasteland of infected rotten corpses. About page 30 there’s a precipating event. The young adults have to go out into the diseased world. They discover there was no plague, few died, average age of the dead 79 and life on earth is better than it was when their idiot parents took them off to the hideout,Replies: @AnotherDad, @jsm, @James N. Kennett
Not the case.
I thought those predictions were over-the-top because the Diamond Princess data established that there were a lot of asymptomatic infections. So the real death rate would be lower than what was reported.
But if that was not true (and you had a full 1% IFR) and this thing exploded–no reaction, no social distancing, no lockdown, no masks–then a couple million dead before getting to herd immunity wasn’t ridiculous at all.
But that doesn’t mean it was ever going to be an apocalypse. 2.8 million Americans–mostly very old–die every year. This thing killing another 2 or even 3 or 4 million–somewhat younger, but still mostly very old Americans–also would not have been the apocalypse, nor brought on any of the things you mention.
~~
I like your movie script.
However, the real kicker would be that when these kids re-discover the world … they’ll at first think their parents were idiots and run out into the big “modern” world ….
but then find out their little world of hard work is actually healthier:
No fats and tats, better behavior, more love, healthier marriages, higher fertility, more children, a brighter future.
The payoff will be most of them deciding to go back into hideout and resume healthy living.
The Trigger Effect is a 1996 American thriller film written and directed by David Koepp and starring Kyle MacLachlan, Elisabeth Shue and Dermot Mulroney. The film follows the downward spiral of society during a widespread and lengthy power outage in Southern California. As Koepp's directorial debut, the film was inspired by the 1978 documentary television series Connections and the 1960 The Twilight Zone episode "The Monsters Are Due on Maple Street", where Koepp's uncle, actor Claude Akins, starred.
Yuck. This is the modern British establishment?
Ferguson an empty man.
His wife with an extremely unpleasant dikey look–the antithesis of feminine–of the sort that would cause any man to loose interest. Zero effort (desire?) to even try and keep her husband.
The slut–an “activist” slut.
Her husband from a Tory family … who the hell is ok with his “wife” out banging other men? Yep, he looks like a pathetic soy boy cuck.
These are really the same people–like cuck boy’s Spitfire flying grandfather–who held fast against the Nazis?
Too bad this virus could finish off the lot of them.
Ferguson himself might have been thinking along the lines of Dr Strangelove where he suggests that key officials be supplied with multiple fertile women to help repopulate the human race.Replies: @O'Really, @Anonymous, @Jenner Ickham Errican, @Mr. Anon, @Alexander Turok, @Bardon Kaldian
We may have some fun with it, but, ultimately ….. whoring, especially female promiscuity, signals the end of a functioning modern, mostly restrained civilization.
Here's her more typical appearance in daily life
https://i.ibb.co/zfrqHHM/antonia-staats-activist.jpgReplies: @George, @Known Fact, @Kent Nationalist
That first face-shot they released made her look like Cate Blanchett
If Biden was British, he could get away with groping or even impregnating a staffer or two, just so long as he did not break quarantine.
Seriously though, it looks like this Reade woman is a liar, but no one wants to say so, because it looks so bad calling a woman who alleges sexual assault a liar.
However, when the man is running for POTUS, that is a bit different from a story about the average bit of slap-and-tickle in the corridors of Congress, where legislators finger some readily available lobbyist vagina in between hearings for the Supreme Court or deciding the future of health care, is it not?
According to Worldometer, Britain’s death rate (451 per million population) remains lower than Italy’s (495/million), as well as lower than Spain’s (558/million), Andorra’s (595/million), Belgium’s (726/million), and San Marino’s (1,208/million).
https://www.worldometers.info/coronavirus/#countries
I wonder if she counts as a mistress if Ferguson isn’t paying for her apartment and clothes.
Whoring and slutting are two different things. A whore is in the business of selling it. A slut is her worst enemy, devaluing her product. Prostitution was a far more lucrative endeavor before open promiscuity became acceptable to Middle America. The very wealthy, the very poor and the artistic/showbiz/counterculture always did as they liked sexually but were circumspect in public about it. Middle class Middle America had much less “illicit” sex but much of that was paid for.
The best refutation of a Ron Unz comment is his website.
Here's her more typical appearance in daily life
https://i.ibb.co/zfrqHHM/antonia-staats-activist.jpgReplies: @George, @Known Fact, @Kent Nationalist
She’s still like a supermodel in comparison with his wife
I’m thinking of a movie script. A group of idiots who believe the NYSLIMES and the rest of the media bug out to some prepper hideout. The hideout was well organized in advance for some end of the world event such as the 2019/20 flu season.
For 30 years the idiots grow their own food, get around on bikes and run their happy little self sufficient society. The children grew up believing the world outside is a vast wasteland of infected rotten corpses. About page 30 there’s a precipating event. The young adults have to go out into the diseased world. They discover there was no plague, few died, average age of the dead 79 and life on earth is better than it was when their idiot parents took them off to the hideout,Replies: @AnotherDad, @jsm, @James N. Kennett
Sorry, buddy, that movie has already been made,
Blast from the Past starring Brendan Fraser. Just, the apocalypse was nuclear war that didn’t happen
I thought those predictions were over-the-top because the Diamond Princess data established that there were a lot of asymptomatic infections. So the real death rate would be lower than what was reported.
But if that was not true (and you had a full 1% IFR) and this thing exploded--no reaction, no social distancing, no lockdown, no masks--then a couple million dead before getting to herd immunity wasn't ridiculous at all.
But that doesn't mean it was ever going to be an apocalypse. 2.8 million Americans--mostly very old--die every year. This thing killing another 2 or even 3 or 4 million--somewhat younger, but still mostly very old Americans--also would not have been the apocalypse, nor brought on any of the things you mention.
~~
I like your movie script.
However, the real kicker would be that when these kids re-discover the world ... they'll at first think their parents were idiots and run out into the big "modern" world ....
but then find out their little world of hard work is actually healthier:
No fats and tats, better behavior, more love, healthier marriages, higher fertility, more children, a brighter future.
The payoff will be most of them deciding to go back into hideout and resume healthy living.Replies: @Clyde, @TomSchmidt
Go see “Trigger Effect” if you don’t know how to “acquire” this then you are plain ass lost
The Trigger Effect is a 1996 American thriller film written and directed by David Koepp and starring Kyle MacLachlan, Elisabeth Shue and Dermot Mulroney. The film follows the downward spiral of society during a widespread and lengthy power outage in Southern California. As Koepp’s directorial debut, the film was inspired by the 1978 documentary television series Connections and the 1960 The Twilight Zone episode “The Monsters Are Due on Maple Street”, where Koepp’s uncle, actor Claude Akins, starred.
But if you're going to simulate, your assumptions should be part of any consideration of your output. And reusing old code is definitely a risk.
The first Ariane 5 rockets blew itself up because its Ariane 4 software didn't like to accelerate so fast.Replies: @Polynikes, @JosephB
I’m a bit more pessimistic about the code.
The author raised the concern that the outputs vary across runs even with identical inputs and with the same random number seed. That is a bit disturbing. Either there is some other source of randomness in the code (e.g., using 2 different libraries for random numbers), or there is something going on with memory manage. Given that the code gives different results when run on multiple cores vs. a single core, I’m guessing race conditions. At a minimum, there is something sloppy going on.
I thought those predictions were over-the-top because the Diamond Princess data established that there were a lot of asymptomatic infections. So the real death rate would be lower than what was reported.
But if that was not true (and you had a full 1% IFR) and this thing exploded--no reaction, no social distancing, no lockdown, no masks--then a couple million dead before getting to herd immunity wasn't ridiculous at all.
But that doesn't mean it was ever going to be an apocalypse. 2.8 million Americans--mostly very old--die every year. This thing killing another 2 or even 3 or 4 million--somewhat younger, but still mostly very old Americans--also would not have been the apocalypse, nor brought on any of the things you mention.
~~
I like your movie script.
However, the real kicker would be that when these kids re-discover the world ... they'll at first think their parents were idiots and run out into the big "modern" world ....
but then find out their little world of hard work is actually healthier:
No fats and tats, better behavior, more love, healthier marriages, higher fertility, more children, a brighter future.
The payoff will be most of them deciding to go back into hideout and resume healthy living.Replies: @Clyde, @TomSchmidt
That was the premise of M. night Shyamalan’s The Village. Not great, not bad, but worth a watch.
I don’t know anything about coding, but it doesn’t sound great for Mr. Fuergeson.Replies: @vhrm, @David, @Jim Don Bob, @James N. Kennett
I worked on models for over 10 years and I though the criticisms were valid. Especially worrisome was that there was no repeatability using the same inputs. This model would not be tolerated in a commercial environment.
Oh, don’t worry.
Time Magazine has the cover story all ready to go . . .
She’s Cate’s double.
those are pretty normal British names, if you were trying to imply something about that?
he needed to drill deeper into the staats.
When your team claims to be specialised in a specific type of modelling, and has a 'base' model that is claimed to be state of the art, then that motherfucker had better be able to be picked up by an numerate layman and understood directly from the documentation.
I would go further: such a 'shop' ought to have a 'toy' version of their model which gives the aforementioned numerate layman a way to understand the stylised facts of the 'big' model.
That's been the 'house view' at CoPS since the 1980s. I was there in the 1990s as an RA and PhD student.
As a very good example: go to the webpage for the very large number of variants of ORANI-G; download any of the ZIP files, and examine any of the TAB files and look how structured it is.
Then look at the DOC file in the package and you will get a full explanation of each of the models equations and variables.
COMPLETE documentation. NO code-spaghetti.
That is the standard expected from a professional group of academic modellers with a 40-year history of global leadership in their discipline. It is at it was when I joined; it was the same when I left in 2000, and it's the same today.
That said: the ICL model code isn't that badly structured. I've seen much much worse - in quant finance there is a very large amount of badly-structured, badly-documented code, that sometimes doesn't even do what it's supposed to (it passes rudimentary tests by accident as a result of some corner case, and then is viewed as 'correct' and then goes into production).
Take a look at this single equation from a MATLAB model that purported to calculate the sector-level returns on a portfolio... note that the red apostrophe indicates a transpose, and the text in blue is effectively a subscript (it's the range - from 1 to the maximum months of data available for the first benchmark for the fund). Originally there were no spaces in the line -
https://www.dropbox.com/s/anbogp8ah3psxwj/ARBS.png?dl=1
Turns out that, both of the quantities on the RHS were already calculated elsewhere (again, on the fly) and were never stored. This happened no less than 8 times in the code, sometimes using different names for the same variables.
So after refactoring, and declaring a convenience variable, the above shitshow becomesAnyhow... refactoring bad code is also a terrific way to work out what it does - and compare that to what the code's owners think it should do.
.
My disagreement with the ICL model is not its stylistic shortcomings; it's the lack of sensitivity analysis that was performed before Ferguson started telling the political class a horror story that jibed with their desire to be seen to save the world. It's the same shit he has pulled roughly twice a decade since the mid-90s - which indicates it wasn't an accident.Replies: @jbwilson24
“Anyhow… refactoring bad code is also a terrific way to work out what it does – and compare that to what the code’s owners think it should do.”
There’s bad code, then there is complete and utter professional negligence. This model was relied upon for actual policy decisions that impact lives, making it safety-critical.
None of the usual software engineering techniques (circa 2000) appear to have been followed. Unit tests, proper modularization, separation of concerns, etc. He wrote this model in 2007, when Java, XUnit, execution tracers and formal verification were around. There’s absolutely no excuse for this complete and utter quackery.
Amateur programmers shouldn’t be anywhere near C or any other language that allows direct access to memory. Ever. Java was widely supported back in 2007, and there were numerous simulation toolkits at an appropriate level of abstraction.
Safety-critical software should be treated with extra care. If you look at other forms of modeling, like transport or nuclear, the various modeling groups typically have actual software engineers on staff. That is, people whose entire skillset involves software. Given that the guy received 130 million or so, he surely could have spent some cash on a couple of actual software engineers.
Unless you have worked for a proper software development company where you’d get yelled at (or poorly reviewed) for bad software technique, you are an amateur. Software is a craft that requires discipline, and you learn by being immersed in an environment where you are guided/smacked by more experienced developers. Amateurs need to take extra care not to do stupid things, but this guy seems to have reveled in his incompetence.
(I won’t even mention other modeling issues, such as various forms of uncertainty and model error that are widely discussed in proper fields like climate or fluid modeling).
He should face discipline for incompetence. The episode with his lover is of little consequence in comparison.
it just highlights how coding implementation is NOT the issue in either of these fields.It's high sensitivity to currently unknown constants.Replies: @anon
Then you run the model, and get statistically-defensible values for 'input' variables that you know will generate the output you want.Swap back to a forecast closure, and run the model... and you get your desired output paths. I discontinued my PhD candidature a month later; if I was going to be a charlatan who used numerical models to add an undeserved veneer to my own priors, I may as well do it in a relatively highly-paid field (quant finance) rather than a low-paid (academic research). Even nowadays, I will only do 'ex post' quant (e.g., performance and risk attribution, and Monte Carlo based stress tests and 'what-if' scenario and event studies), and I refuse to do numerical forecasts unless the results explicitly include a full, systematic sensitivity analysis. There's no real market for that anyway: people want lines, not 3D surfaces. At least lines are more informative than how things were in 1995 when I did dozens of analyses of tax-mix change for PwC corporate clients... they just wanted 'X%' - the long-run percentage change in the total cost for the client.It's weird to be actively hostile to numerical forecasting (as currently practised), considering that my PhD topic was "Forward-Looking Portfolio Analysis via Stochastic Simulation in a Dynamic Multisector CGE Model" (or some similar word-salad).Third (after the 'More'): some of the CompSci critiques are outright wrong - and show why all the CompSci 'best practice' in the world would not make a dime's worth of difference. Almost all of them represent CompSci people trying to flex on a technical discipline, because CompSci is institutionally insecure.This in turn is because outside the top couple of percent, most coders are shit: after all, this is the métier responsible for releasing every 0Day vulnerability into the wild; for every enterprise data breach, and - even worse - for Windows Update). For example... someone had it in their pretty little CompSci head that two model runs with the same inputs should produce identical outputs. That's just no so, in general: it's a problem in the ICL model because it wasn't what they were going for. So the code was bad because it didn't do what ICL thought it was doing - not because the model did not 'replicate'. My models are generally not one-shot replicable either, because some of the parameters are used to parameterise distributions which generate transition matrices that change (quasi-)randomly at each point in forecast time..Fourth: What is to be done?Top-flight research groups in technical disciplines, need to be able to write and maintain their own code as part of a coherent project strategy.This does not require a Full-stack Research Software Engineer Science Analyst Jedi Mentat Architect Ninja Ultra JSOC Power Pro Plus Level Infinity (which is what some insecure CompSci retard will want on their business card - in the same way that the senior HR fuckwit at a major law firm will want to be the 'HR Partner' despite not being involved in the partnership itself). What it does require is that new inductees into the research group - all of whom will have significant training in whatever form of statistical method is being used - are shown the house style, and are encouraged to maintain their own code according to that style.I always refer to CoPS (where I was a grad student and Junior Minion in the 90s) is an absolute benchmark for exceptional planet-wide excellence in numerical modelling; its reputation in its discipline is unrivalled. For the vast bulk of its existence it had zero 'professional' developers. I think it has one or two now, because the team has gone from ~10 + admin when I was a Junior Minion there.The main solution software for CoPS-style models (called GEMPACK) is used by pretty much every CGE research group in the world, and it was developed, virtually solo, by a mathematician (Ken Pearson). To quote from a tribute written in 2015 (when Ken died):The other 'main guy' for GEMPACK is an economist (Mark Horridge).[1] CoPS was already so well-regarded in the field at that time, that nobody had to sink to the 'guns for hire' level. In fact they were able to choose their contract research so that they could answer things that they thought were interesting. It was clear, though, that academia was declining so fast that even a place as great as CoPS faced interminable interference from university administrators - the senior faculty could tell them to fuck off, but I would have been a lowly postdoc.Replies: @Steve Sailer, @res, @anon
I don’t know anything about coding, but it doesn’t sound great for Mr. Fuergeson.Replies: @vhrm, @David, @Jim Don Bob, @James N. Kennett
It’s pretty bad – and this is the code that has been cleaned up by a team at Microsoft. Perhaps one day a disgruntled student will leak the original C code for us to marvel at.
The quality of the code will be hugely embarrassing for Imperial College, and would not have been acceptable in the government departments that Ferguson was advising – they adopted coding standards decades ago.
Whether the output of the code is trustworthy is another matter.
https://twitter.com/JoshSchoen/status/1257756595859140608Replies: @Neoconned, @EdwardM
…..and yet he keeps getting hired…..
I’ve worked in casinos and restaurants for most of the last decade or so….with as much as they ride my ass if I had performed like that I wouldn’t have a job now…
I’m thinking of a movie script. A group of idiots who believe the NYSLIMES and the rest of the media bug out to some prepper hideout. The hideout was well organized in advance for some end of the world event such as the 2019/20 flu season.
For 30 years the idiots grow their own food, get around on bikes and run their happy little self sufficient society. The children grew up believing the world outside is a vast wasteland of infected rotten corpses. About page 30 there’s a precipating event. The young adults have to go out into the diseased world. They discover there was no plague, few died, average age of the dead 79 and life on earth is better than it was when their idiot parents took them off to the hideout,Replies: @AnotherDad, @jsm, @James N. Kennett
Nope. Ferguson never predicted the death of more than 1% of the population.
https://twitter.com/JoshSchoen/status/1257756595859140608Replies: @Neoconned, @EdwardM
Hey, in 2002 he “Predicted that between 50 and 50,000 people could [sic] die” of mad cow disease. The toll ended up being around 200. He was right!
There is obvioulsy a great market for credentialed alarmists. Even more so in the UK, the ultimate nanny state, than in the U.S.
There's bad code, then there is complete and utter professional negligence. This model was relied upon for actual policy decisions that impact lives, making it safety-critical.
None of the usual software engineering techniques (circa 2000) appear to have been followed. Unit tests, proper modularization, separation of concerns, etc. He wrote this model in 2007, when Java, XUnit, execution tracers and formal verification were around. There's absolutely no excuse for this complete and utter quackery.
Amateur programmers shouldn't be anywhere near C or any other language that allows direct access to memory. Ever. Java was widely supported back in 2007, and there were numerous simulation toolkits at an appropriate level of abstraction.
Safety-critical software should be treated with extra care. If you look at other forms of modeling, like transport or nuclear, the various modeling groups typically have actual software engineers on staff. That is, people whose entire skillset involves software. Given that the guy received 130 million or so, he surely could have spent some cash on a couple of actual software engineers.
Unless you have worked for a proper software development company where you'd get yelled at (or poorly reviewed) for bad software technique, you are an amateur. Software is a craft that requires discipline, and you learn by being immersed in an environment where you are guided/smacked by more experienced developers. Amateurs need to take extra care not to do stupid things, but this guy seems to have reveled in his incompetence.
(I won't even mention other modeling issues, such as various forms of uncertainty and model error that are widely discussed in proper fields like climate or fluid modeling).
He should face discipline for incompetence. The episode with his lover is of little consequence in comparison.Replies: @vhrm, @Kratoklastes
This isn’t safety-critical anything. It’s just random academic code.
Should it be better? yes, but let’s not get the vapors over it either.
And there’s no evidence or suggestion that its undocumented, single-threaded, non-repeatable behavior had any relevant bearing on the results it gave in this case.
also:
It’s interesting you would bring this up. Taking your word that these are bullet-proof, well abstracted, well componentized, well documented, formally verified (though i HIGHLY doubt this one), unit tested to 110% coverage marvels of software engineering…
it just highlights how coding implementation is NOT the issue in either of these fields.
It’s high sensitivity to currently unknown constants.
Should it be better? yes, but let’s not get the vapors over it either.Isn't this the simulation code that produced the "2 million dead Americans" result?
If so, this simulation code was used to formulate policy. That makes it critical. And there’s no evidence or suggestion that its undocumented, single-threaded, non-repeatable behavior had any relevant bearing on the results it gave in this case.That's just a dumb thing to write. I'll try to explain via analogy. In your modern car there is no direct connection between the accelerator pedal and the engine. Your accelerator pedal is just like a joystick on a Playstation, it makes suggestions to the engine / car controller. Suppose that sometimes when you press on the accelerator the engine revved nicely and the car accelerated briskly, but other times the engine dropped to idle and stayed there. Would you take that car on a crowded freeway? Repeatability is part of testability, in fact it is part of the basic scientific method. An experiment that isn't repeatable may be interesting but it ain't science. tl;dr
This code is junk and should never have been involved in any actual real-world decision making. It cannot be defended as a policy tool.
Riken Patel, Kenyan Gujirati Indian father suggest he might be Jain, Jains are the worst, the Jews of Indian.
Does sound a little like yet another Indian cult.
Clear ties to the Zionist lobby too, with J Street and the foreign policy stances they have advocated.
For you Reg, to be refined as you see fit:
So an historian, a cryptographer, and an epidemiologist go duck hunting.
A duck rises from the lake. The cryptographer jerks his gun up and fires 10 feet in front of the duck. “I was sure that’s where it was going” he murmurs, disappointed.
The historian takes slow, careful aim, tracking the duck over a quarter of the sky, fires and misses ten feet behind the duck. “I’m quite positive that’s where it was,” he says dejectedly.
The epidemiologist watches the duck fly off and says quietly
“I’m sure we hit it once.”
it just highlights how coding implementation is NOT the issue in either of these fields.It's high sensitivity to currently unknown constants.Replies: @anon
This isn’t safety-critical anything. It’s just random academic code.
Should it be better? yes, but let’s not get the vapors over it either.
Isn’t this the simulation code that produced the “2 million dead Americans” result?
If so, this simulation code was used to formulate policy. That makes it critical.
And there’s no evidence or suggestion that its undocumented, single-threaded, non-repeatable behavior had any relevant bearing on the results it gave in this case.
That’s just a dumb thing to write. I’ll try to explain via analogy.
In your modern car there is no direct connection between the accelerator pedal and the engine. Your accelerator pedal is just like a joystick on a Playstation, it makes suggestions to the engine / car controller. Suppose that sometimes when you press on the accelerator the engine revved nicely and the car accelerated briskly, but other times the engine dropped to idle and stayed there. Would you take that car on a crowded freeway?
Repeatability is part of testability, in fact it is part of the basic scientific method. An experiment that isn’t repeatable may be interesting but it ain’t science.
tl;dr
This code is junk and should never have been involved in any actual real-world decision making. It cannot be defended as a policy tool.
For example, the IHME model could get hard data on day to day hospital admissions which was part of their chart. They would release updates that were off by as much as 50% for their next day forecasts. That is almost impossible. A child drawing a straight line on the graph could've had more accurate guesses. Similarly, Wisconsin released one of their projections that supposedly justifies the shutdown; it had projections for dates already passed that were up to 100% off. In other words, they couldn't separate themselves from their theory to even reconcile with the reality in front of their face.That article is mostly foreign to me, because I know nothing about coding. Do you get the impression Furgeson was sloppy, stupid, malicious, or some combination thereof?Replies: @David
I’m not sure. Models need to be used and compared to actual outcomes a lot before it makes any sense to rely on them. I would have thought that modeling how a disease propagates though a population would be pretty well worked out. I agree there’s a strange reluctance to compare projections to actual results, and to adjust going-forward projections as a result.
In a college finance text book, I read that if a year into your business plan, you’re at half your original revenue target, you should halve all your projected revenues. I’ve never seen a company do that.
Agreed. The author says that deterministic steps had been replaced with probabilistic ones. Like, say you come to believe that smokers are 40% less likely to get infected, and that they make up 20% of the population. You might take what had been a fixed chance of getting infected of 4.6% and replace it with a 20% chance of 3% infection rate and a 80% chance of 5% infection rate.
So what began as a deterministic model run with various assumptions on a fixed set of random inputs to gauge the impact of different assumptions, gained one or more levels of randomness, obscuring the impact of any change that was tested.
Likely, rather than come up with more “pseudo-random” inputs, they just averaged multiple outcomes to try to dampen out the new volatility.
But I bet the biggest problem with the model was the assumption about fatality, which isn’t really a problem with the model at all.
There's bad code, then there is complete and utter professional negligence. This model was relied upon for actual policy decisions that impact lives, making it safety-critical.
None of the usual software engineering techniques (circa 2000) appear to have been followed. Unit tests, proper modularization, separation of concerns, etc. He wrote this model in 2007, when Java, XUnit, execution tracers and formal verification were around. There's absolutely no excuse for this complete and utter quackery.
Amateur programmers shouldn't be anywhere near C or any other language that allows direct access to memory. Ever. Java was widely supported back in 2007, and there were numerous simulation toolkits at an appropriate level of abstraction.
Safety-critical software should be treated with extra care. If you look at other forms of modeling, like transport or nuclear, the various modeling groups typically have actual software engineers on staff. That is, people whose entire skillset involves software. Given that the guy received 130 million or so, he surely could have spent some cash on a couple of actual software engineers.
Unless you have worked for a proper software development company where you'd get yelled at (or poorly reviewed) for bad software technique, you are an amateur. Software is a craft that requires discipline, and you learn by being immersed in an environment where you are guided/smacked by more experienced developers. Amateurs need to take extra care not to do stupid things, but this guy seems to have reveled in his incompetence.
(I won't even mention other modeling issues, such as various forms of uncertainty and model error that are widely discussed in proper fields like climate or fluid modeling).
He should face discipline for incompetence. The episode with his lover is of little consequence in comparison.Replies: @vhrm, @Kratoklastes
Firstly: climate modelling is not a ‘proper’ field (hint: Ferguson’s balls deep in that gigantic vortex of charlatanry that pays top dollar for anyone willing to reliably forecast catastrophe).
Second: the core problem is not one of layout or testing or DRY principles, or stuff like that. All of those are good ideas and should form part of good research practice – but good research practice is irrelevant if your aim is to stick a vacuum hose into a bucket of government money, when getting to do so requires you to tell a specific story.
So the key issue is not the C++ chops or git-fu of the developers; it’s the existence of (or absence of) a coherent structure in the research process ‘deep down’, and the overarching objectives of the research leaders.
Normal ‘software project’ stuff (code repo/SVN, directory structure, modularity, unit tests, DRY, documentation) – all of which are perfectly reasonable in and of themselves if they’re used properly – aren’t going to help when the entire research structure is geared around opportunistically crying “Wolf!!” whenever anyone finds a new pathogen.
The core problem with the ICL model is that the people running it – Ferguson especially – are catastrophists who want to tell a story, and they ‘know’ their model.
If a numerical modeller knows the model, then he knows how to produce whatever output he wants, while keeping all inputs within statistically-defensible ranges.
Any quant who genuinely understands their model, knows how to change the parameters (and exogenous variables) to produce whatever path they want – without moving the parameters outside of an envelope that passes statistical scrutiny (e.g., an ML test would say that the parameter set as a whole was not statistically different from its ML values).
And ‘peer review’ – even if it was conducted by the most dispassionate reviewer – won’t help.
In a model of even slight complexity, no reviewer is going to examine every input (which will include forecast paths for regressors) if every input lies within a 0.5σ band of its estimate and all paths ‘look about right’ (i.e., they have defensible short run paths and converge to a stable equilibrium path).
This means that Gresham’s Law even applies to numerical forecasters: ‘bad’ forecasts will drive out good.
To this day I know the precise date that it became clear that the modelling paradigm would devolve into ‘guns for hire’, telling government what it wanted to hear[1]: Friday, 13 August 1999.
That was the day that I finished two little tools (‘Tweaker‘ and ‘Fudger‘) that could automate model-closure swaps to ‘goal seek’ a result for a desired set of outputs, while ensuring that the model inputs were statistically defensible.
It involved changing the model ‘closure’ so that some output paths were made exogenous, and some naturally exogenous (input) variables were made endogenous (i.e., solved by the model). Tweaker & Fudger added a set of additional equations to impose range constraints on the swapped-to-be-endogenous variables (to ensure they were kept within 0.5σ of the base forecast).
Then you run the model, and get statistically-defensible values for ‘input’ variables that you know will generate the output you want.
Swap back to a forecast closure, and run the model… and you get your desired output paths.
I discontinued my PhD candidature a month later; if I was going to be a charlatan who used numerical models to add an undeserved veneer to my own priors, I may as well do it in a relatively highly-paid field (quant finance) rather than a low-paid (academic research).
Even nowadays, I will only do ‘ex post’ quant (e.g., performance and risk attribution, and Monte Carlo based stress tests and ‘what-if‘ scenario and event studies), and I refuse to do numerical forecasts unless the results explicitly include a full, systematic sensitivity analysis. There’s no real market for that anyway: people want lines, not 3D surfaces. At least lines are more informative than how things were in 1995 when I did dozens of analyses of tax-mix change for PwC corporate clients… they just wanted ‘X%’ – the long-run percentage change in the total cost for the client.
It’s weird to be actively hostile to numerical forecasting (as currently practised), considering that my PhD topic was “Forward-Looking Portfolio Analysis via Stochastic Simulation in a Dynamic Multisector CGE Model” (or some similar word-salad).
Third (after the ‘More’): some of the CompSci critiques are outright wrong – and show why all the CompSci ‘best practice’ in the world would not make a dime’s worth of difference.
Almost all of them represent CompSci people trying to flex on a technical discipline, because CompSci is institutionally insecure.This in turn is because outside the top couple of percent, most coders are shit: after all, this is the métier responsible for releasing every 0Day vulnerability into the wild; for every enterprise data breach, and – even worse – for Windows Update).
For example… someone had it in their pretty little CompSci head that two model runs with the same inputs should produce identical outputs.
That’s just no so, in general: it’s a problem in the ICL model because it wasn’t what they were going for.
So the code was bad because it didn’t do what ICL thought it was doing – not because the model did not ‘replicate’. My models are generally not one-shot replicable either, because some of the parameters are used to parameterise distributions which generate transition matrices that change (quasi-)randomly at each point in forecast time.
.
Fourth: What is to be done?
Top-flight research groups in technical disciplines, need to be able to write and maintain their own code as part of a coherent project strategy.
This does not require a Full-stack Research Software Engineer Science Analyst Jedi Mentat Architect Ninja Ultra JSOC Power Pro Plus Level Infinity (which is what some insecure CompSci retard will want on their business card – in the same way that the senior HR fuckwit at a major law firm will want to be the ‘HR Partner‘ despite not being involved in the partnership itself).
What it does require is that new inductees into the research group – all of whom will have significant training in whatever form of statistical method is being used – are shown the house style, and are encouraged to maintain their own code according to that style.
I always refer to CoPS (where I was a grad student and Junior Minion in the 90s) is an absolute benchmark for exceptional planet-wide excellence in numerical modelling; its reputation in its discipline is unrivalled.
For the vast bulk of its existence it had zero ‘professional’ developers. I think it has one or two now, because the team has gone from ~10 + admin when I was a Junior Minion there.
The main solution software for CoPS-style models (called GEMPACK) is used by pretty much every CGE research group in the world, and it was developed, virtually solo, by a mathematician (Ken Pearson). To quote from a tribute written in 2015 (when Ken died):
The other ‘main guy’ for GEMPACK is an economist (Mark Horridge).
[1] CoPS was already so well-regarded in the field at that time, that nobody had to sink to the ‘guns for hire’ level. In fact they were able to choose their contract research so that they could answer things that they thought were interesting.
It was clear, though, that academia was declining so fast that even a place as great as CoPS faced interminable interference from university administrators – the senior faculty could tell them to fuck off, but I would have been a lowly postdoc.
2. Tell them.
3. Tell them what you told them.Replies: @vhrm, @Jim Don Bob
Then you run the model, and get statistically-defensible values for 'input' variables that you know will generate the output you want.Swap back to a forecast closure, and run the model... and you get your desired output paths. I discontinued my PhD candidature a month later; if I was going to be a charlatan who used numerical models to add an undeserved veneer to my own priors, I may as well do it in a relatively highly-paid field (quant finance) rather than a low-paid (academic research). Even nowadays, I will only do 'ex post' quant (e.g., performance and risk attribution, and Monte Carlo based stress tests and 'what-if' scenario and event studies), and I refuse to do numerical forecasts unless the results explicitly include a full, systematic sensitivity analysis. There's no real market for that anyway: people want lines, not 3D surfaces. At least lines are more informative than how things were in 1995 when I did dozens of analyses of tax-mix change for PwC corporate clients... they just wanted 'X%' - the long-run percentage change in the total cost for the client.It's weird to be actively hostile to numerical forecasting (as currently practised), considering that my PhD topic was "Forward-Looking Portfolio Analysis via Stochastic Simulation in a Dynamic Multisector CGE Model" (or some similar word-salad).Third (after the 'More'): some of the CompSci critiques are outright wrong - and show why all the CompSci 'best practice' in the world would not make a dime's worth of difference. Almost all of them represent CompSci people trying to flex on a technical discipline, because CompSci is institutionally insecure.This in turn is because outside the top couple of percent, most coders are shit: after all, this is the métier responsible for releasing every 0Day vulnerability into the wild; for every enterprise data breach, and - even worse - for Windows Update). For example... someone had it in their pretty little CompSci head that two model runs with the same inputs should produce identical outputs. That's just no so, in general: it's a problem in the ICL model because it wasn't what they were going for. So the code was bad because it didn't do what ICL thought it was doing - not because the model did not 'replicate'. My models are generally not one-shot replicable either, because some of the parameters are used to parameterise distributions which generate transition matrices that change (quasi-)randomly at each point in forecast time..Fourth: What is to be done?Top-flight research groups in technical disciplines, need to be able to write and maintain their own code as part of a coherent project strategy.This does not require a Full-stack Research Software Engineer Science Analyst Jedi Mentat Architect Ninja Ultra JSOC Power Pro Plus Level Infinity (which is what some insecure CompSci retard will want on their business card - in the same way that the senior HR fuckwit at a major law firm will want to be the 'HR Partner' despite not being involved in the partnership itself). What it does require is that new inductees into the research group - all of whom will have significant training in whatever form of statistical method is being used - are shown the house style, and are encouraged to maintain their own code according to that style.I always refer to CoPS (where I was a grad student and Junior Minion in the 90s) is an absolute benchmark for exceptional planet-wide excellence in numerical modelling; its reputation in its discipline is unrivalled. For the vast bulk of its existence it had zero 'professional' developers. I think it has one or two now, because the team has gone from ~10 + admin when I was a Junior Minion there.The main solution software for CoPS-style models (called GEMPACK) is used by pretty much every CGE research group in the world, and it was developed, virtually solo, by a mathematician (Ken Pearson). To quote from a tribute written in 2015 (when Ken died):The other 'main guy' for GEMPACK is an economist (Mark Horridge).[1] CoPS was already so well-regarded in the field at that time, that nobody had to sink to the 'guns for hire' level. In fact they were able to choose their contract research so that they could answer things that they thought were interesting. It was clear, though, that academia was declining so fast that even a place as great as CoPS faced interminable interference from university administrators - the senior faculty could tell them to fuck off, but I would have been a lowly postdoc.Replies: @Steve Sailer, @res, @anon
You should try the method taught in junior high school composition:
1. Tell them what you are going to tell them.
2. Tell them.
3. Tell them what you told them.
Sometimes the summary gets its own summary.
And I'm still not sure how much people actually read.Replies: @Chrisnonymous
2. Tell them.
3. Tell them what you told them.Replies: @vhrm, @Jim Don Bob
Heh. last 5 years or so all of my emails that are more than a couple of sentences have a summary at the top.
Sometimes the summary gets its own summary.
And I’m still not sure how much people actually read.
Probably ill-advised to associate J-street with the Zionazis.
Ferguson’s wife looks awful. She has made no effort to make herself sexy or appealing to her not bad looking husband. It is therefore natural that he should look elsewhere, as everyone will eventually walk away from a bad deal. He’s famous after all for better or for worse so he is an obvious magnet for epidemiologist-star groupies.
It is annoying that men from the Anglosphere, who are arguably the most attentive and caring of boyfriends or husbands, get to choose between landwhales or frumps. When I go to the supermarket I see plenty of middle aged men who are often quite presentable, saddled with a wife who is as wide as she is tall. Who could blame these men if they had a bit on the side. I know I would. When I do see a slim decent looking bird she’s invariably an Eastern European and her boyfriend is a pot bellied derelict.
Chateau Heartiste put it best…”If you want to know whether a man is happy in his relationship, then just ask yourself “Would you f**k her ?” If the answer is “yes”, then you know he is happy.”
Then you run the model, and get statistically-defensible values for 'input' variables that you know will generate the output you want.Swap back to a forecast closure, and run the model... and you get your desired output paths. I discontinued my PhD candidature a month later; if I was going to be a charlatan who used numerical models to add an undeserved veneer to my own priors, I may as well do it in a relatively highly-paid field (quant finance) rather than a low-paid (academic research). Even nowadays, I will only do 'ex post' quant (e.g., performance and risk attribution, and Monte Carlo based stress tests and 'what-if' scenario and event studies), and I refuse to do numerical forecasts unless the results explicitly include a full, systematic sensitivity analysis. There's no real market for that anyway: people want lines, not 3D surfaces. At least lines are more informative than how things were in 1995 when I did dozens of analyses of tax-mix change for PwC corporate clients... they just wanted 'X%' - the long-run percentage change in the total cost for the client.It's weird to be actively hostile to numerical forecasting (as currently practised), considering that my PhD topic was "Forward-Looking Portfolio Analysis via Stochastic Simulation in a Dynamic Multisector CGE Model" (or some similar word-salad).Third (after the 'More'): some of the CompSci critiques are outright wrong - and show why all the CompSci 'best practice' in the world would not make a dime's worth of difference. Almost all of them represent CompSci people trying to flex on a technical discipline, because CompSci is institutionally insecure.This in turn is because outside the top couple of percent, most coders are shit: after all, this is the métier responsible for releasing every 0Day vulnerability into the wild; for every enterprise data breach, and - even worse - for Windows Update). For example... someone had it in their pretty little CompSci head that two model runs with the same inputs should produce identical outputs. That's just no so, in general: it's a problem in the ICL model because it wasn't what they were going for. So the code was bad because it didn't do what ICL thought it was doing - not because the model did not 'replicate'. My models are generally not one-shot replicable either, because some of the parameters are used to parameterise distributions which generate transition matrices that change (quasi-)randomly at each point in forecast time..Fourth: What is to be done?Top-flight research groups in technical disciplines, need to be able to write and maintain their own code as part of a coherent project strategy.This does not require a Full-stack Research Software Engineer Science Analyst Jedi Mentat Architect Ninja Ultra JSOC Power Pro Plus Level Infinity (which is what some insecure CompSci retard will want on their business card - in the same way that the senior HR fuckwit at a major law firm will want to be the 'HR Partner' despite not being involved in the partnership itself). What it does require is that new inductees into the research group - all of whom will have significant training in whatever form of statistical method is being used - are shown the house style, and are encouraged to maintain their own code according to that style.I always refer to CoPS (where I was a grad student and Junior Minion in the 90s) is an absolute benchmark for exceptional planet-wide excellence in numerical modelling; its reputation in its discipline is unrivalled. For the vast bulk of its existence it had zero 'professional' developers. I think it has one or two now, because the team has gone from ~10 + admin when I was a Junior Minion there.The main solution software for CoPS-style models (called GEMPACK) is used by pretty much every CGE research group in the world, and it was developed, virtually solo, by a mathematician (Ken Pearson). To quote from a tribute written in 2015 (when Ken died):The other 'main guy' for GEMPACK is an economist (Mark Horridge).[1] CoPS was already so well-regarded in the field at that time, that nobody had to sink to the 'guns for hire' level. In fact they were able to choose their contract research so that they could answer things that they thought were interesting. It was clear, though, that academia was declining so fast that even a place as great as CoPS faced interminable interference from university administrators - the senior faculty could tell them to fuck off, but I would have been a lowly postdoc.Replies: @Steve Sailer, @res, @anon
You forgot the small caveat of “given the same seed for the random number generator.” Given that, do you disagree that the results should be the same? If so, then how do you create tests to check if code changes have created unexpected changes in the output?
It can be turned into a formal input - using random.seed in Python, or set.seed in R - which is very useful for testing.
Semantically, the seed's not part of the model (or its inputs); it's an attempt to replicate randomness so that some variable can have deliberate 'fuzziness'.
The actual input is some variable X which has a Beta(α,β) distribution (say), and the seed is a way to get a drawing from that distribution given α and β.
Under normal circumstances the seed is an output - it is not 'given' to the model, but is generated each time a drawing is made from the distribution. If those were saved, then it would be clearer why the model produced different output for the same inputs.
In strictly numerical terms it's true if the same seed was used to generate a random number from a Beta(α,β) distribution, then the number generated would be the same.That's where set.seed() (or random.seed()) come in.
In a 'test' closure[1] the seed is made exogenous to the model, by switching on an equation for it that effectively switches off the assumed "inherent" randomness.
When the 'seed' equations are switched on, everything's back to being deterministic and normal unit-testing works.
[1] A closure is a partition of the variables in the model into 'given to the model' (exogenous) and 'determined within the model' (endogenous).
Models always has more variables (N) than equations (M > N), and so (M-N) values have to be given in order for the model to solve.
Selection of the (M-N) exogenous variables changes how the model 'works' (what things it is determining).
There are at least 5 different closures that are routine -
• testing* (outputs: depends);
• estimation (outputs: parameters and residuals);
• historical validation (outputs: 'naturally' endogenous variables);
• forecast (outputs: 'naturally' endogenous variables).
Historical validation closures are when you discover how bad a forecast your model would have produced if you ran the model in forecast mode, giving it exactly-correct values for the historical values used for estimation.
Testing has an asterisk because it is really a quasi-closure, because it adds equations that aren't really part of the model. It's only used to check that the code is correctly implemented, and has to happen before any estimation is done.Replies: @res
2. Tell them.
3. Tell them what you told them.Replies: @vhrm, @Jim Don Bob
You do the same thing when you are giving a speech.
It is annoying that men from the Anglosphere, who are arguably the most attentive and caring of boyfriends or husbands, get to choose between landwhales or frumps. When I go to the supermarket I see plenty of middle aged men who are often quite presentable, saddled with a wife who is as wide as she is tall. Who could blame these men if they had a bit on the side. I know I would. When I do see a slim decent looking bird she's invariably an Eastern European and her boyfriend is a pot bellied derelict.
Chateau Heartiste put it best..."If you want to know whether a man is happy in his relationship, then just ask yourself "Would you f**k her ?" If the answer is "yes", then you know he is happy."Replies: @Jim Don Bob
What happened to Heartiste? I enjoyed his style of invective but I’ve seen no activity lately.
Then you run the model, and get statistically-defensible values for 'input' variables that you know will generate the output you want.Swap back to a forecast closure, and run the model... and you get your desired output paths. I discontinued my PhD candidature a month later; if I was going to be a charlatan who used numerical models to add an undeserved veneer to my own priors, I may as well do it in a relatively highly-paid field (quant finance) rather than a low-paid (academic research). Even nowadays, I will only do 'ex post' quant (e.g., performance and risk attribution, and Monte Carlo based stress tests and 'what-if' scenario and event studies), and I refuse to do numerical forecasts unless the results explicitly include a full, systematic sensitivity analysis. There's no real market for that anyway: people want lines, not 3D surfaces. At least lines are more informative than how things were in 1995 when I did dozens of analyses of tax-mix change for PwC corporate clients... they just wanted 'X%' - the long-run percentage change in the total cost for the client.It's weird to be actively hostile to numerical forecasting (as currently practised), considering that my PhD topic was "Forward-Looking Portfolio Analysis via Stochastic Simulation in a Dynamic Multisector CGE Model" (or some similar word-salad).Third (after the 'More'): some of the CompSci critiques are outright wrong - and show why all the CompSci 'best practice' in the world would not make a dime's worth of difference. Almost all of them represent CompSci people trying to flex on a technical discipline, because CompSci is institutionally insecure.This in turn is because outside the top couple of percent, most coders are shit: after all, this is the métier responsible for releasing every 0Day vulnerability into the wild; for every enterprise data breach, and - even worse - for Windows Update). For example... someone had it in their pretty little CompSci head that two model runs with the same inputs should produce identical outputs. That's just no so, in general: it's a problem in the ICL model because it wasn't what they were going for. So the code was bad because it didn't do what ICL thought it was doing - not because the model did not 'replicate'. My models are generally not one-shot replicable either, because some of the parameters are used to parameterise distributions which generate transition matrices that change (quasi-)randomly at each point in forecast time..Fourth: What is to be done?Top-flight research groups in technical disciplines, need to be able to write and maintain their own code as part of a coherent project strategy.This does not require a Full-stack Research Software Engineer Science Analyst Jedi Mentat Architect Ninja Ultra JSOC Power Pro Plus Level Infinity (which is what some insecure CompSci retard will want on their business card - in the same way that the senior HR fuckwit at a major law firm will want to be the 'HR Partner' despite not being involved in the partnership itself). What it does require is that new inductees into the research group - all of whom will have significant training in whatever form of statistical method is being used - are shown the house style, and are encouraged to maintain their own code according to that style.I always refer to CoPS (where I was a grad student and Junior Minion in the 90s) is an absolute benchmark for exceptional planet-wide excellence in numerical modelling; its reputation in its discipline is unrivalled. For the vast bulk of its existence it had zero 'professional' developers. I think it has one or two now, because the team has gone from ~10 + admin when I was a Junior Minion there.The main solution software for CoPS-style models (called GEMPACK) is used by pretty much every CGE research group in the world, and it was developed, virtually solo, by a mathematician (Ken Pearson). To quote from a tribute written in 2015 (when Ken died):The other 'main guy' for GEMPACK is an economist (Mark Horridge).[1] CoPS was already so well-regarded in the field at that time, that nobody had to sink to the 'guns for hire' level. In fact they were able to choose their contract research so that they could answer things that they thought were interesting. It was clear, though, that academia was declining so fast that even a place as great as CoPS faced interminable interference from university administrators - the senior faculty could tell them to fuck off, but I would have been a lowly postdoc.Replies: @Steve Sailer, @res, @anon
For example… someone had it in their pretty little CompSci head that two model runs with the same inputs should produce identical outputs.
You should read more carefully what was actually being tested.
Plus there is the small problem of platform dependence; the same code run on different platforms providing different results. This could be a compiler issue, it could be a memory hierarchy issue, it could be accumulated rounding issues, etc. but it is a red flag in the world of science for the obvious reason of “replication”.
“Run the simulation a bunch of times and average the outputs” isn’t even acceptable at the undergrad level. It’s utterly unprofessional, bordering on academic fraud, at the post grad level.
It’s insanity at the “determine national policy” level.
No, I don’t disagree – except that the seed’s not a “formal” input.
It can be turned into a formal input – using random.seed in Python, or set.seed in R – which is very useful for testing.
Semantically, the seed’s not part of the model (or its inputs); it’s an attempt to replicate randomness so that some variable can have deliberate ‘fuzziness’.
The actual input is some variable X which has a Beta(α,β) distribution (say), and the seed is a way to get a drawing from that distribution given α and β.
Under normal circumstances the seed is an output – it is not ‘given’ to the model, but is generated each time a drawing is made from the distribution. If those were saved, then it would be clearer why the model produced different output for the same inputs.
In strictly numerical terms it’s true if the same seed was used to generate a random number from a Beta(α,β) distribution, then the number generated would be the same.
That’s where set.seed() (or random.seed()) come in.
In a ‘test’ closure[1] the seed is made exogenous to the model, by switching on an equation for it that effectively switches off the assumed “inherent” randomness.
When the ‘seed’ equations are switched on, everything’s back to being deterministic and normal unit-testing works.
[1] A closure is a partition of the variables in the model into ‘given to the model‘ (exogenous) and ‘determined within the model‘ (endogenous).
Models always has more variables (N) than equations (M > N), and so (M–N) values have to be given in order for the model to solve.
Selection of the (M-N) exogenous variables changes how the model ‘works’ (what things it is determining).
There are at least 5 different closures that are routine –
• testing* (outputs: depends);
• estimation (outputs: parameters and residuals);
• historical validation (outputs: ‘naturally’ endogenous variables);
• forecast (outputs: ‘naturally’ endogenous variables).
Historical validation closures are when you discover how bad a forecast your model would have produced if you ran the model in forecast mode, giving it exactly-correct values for the historical values used for estimation.
Testing has an asterisk because it is really a quasi-closure, because it adds equations that aren’t really part of the model. It’s only used to check that the code is correctly implemented, and has to happen before any estimation is done.
https://www.armstrongeconomics.com/world-news/corruption/i-have-reviewed-fergusons-code/
"However, we haven’t had the time to work out a scalable and maintainable way of running the regression test in a way that allows a small amount of variation, but doesn’t let the figures drift over time."As for whether the seed is an input. It is an input to the code which is how I think of this. And BTW, if you are counting the seed as an exogenous variable then I think you need to give the model M-N+1 variables (including the seed as the 1) since it does not appear explicitly in the model equations.Perhaps it would be more accurate to say the seed is both an input and an output. The starting seed is an input and the resulting seed after any given number of calls to the pseudo-random number generator have been made is an output. The starting seed is generally derived from some varying system value (e.g. time) if left unspecified.Your use of "closure" seems nonstandard to me. I am used to definitions along these lines:
https://softwareengineering.stackexchange.com/questions/40454/what-is-a-closure
I mostly get what you mean from context (it's similar to, but not quite the same as, the usage I am used to), but a reference that describes your usage would be helpful.The decomposition into estimation, historical validation, and forecast is helpful. I am quite curious about what historical validation they have done. Especially given Ferguson's poor prediction record in the past.P.S. Your comment reads like you think I did not know about set.seed etc. Please don't assume I am clueless. Also, your comment reads like it is tending towards complicated terminology for the sake of obfuscation rather than clarification. I consider that a bug, not a feature.Replies: @Kratoklastes
It can be turned into a formal input - using random.seed in Python, or set.seed in R - which is very useful for testing.
Semantically, the seed's not part of the model (or its inputs); it's an attempt to replicate randomness so that some variable can have deliberate 'fuzziness'.
The actual input is some variable X which has a Beta(α,β) distribution (say), and the seed is a way to get a drawing from that distribution given α and β.
Under normal circumstances the seed is an output - it is not 'given' to the model, but is generated each time a drawing is made from the distribution. If those were saved, then it would be clearer why the model produced different output for the same inputs.
In strictly numerical terms it's true if the same seed was used to generate a random number from a Beta(α,β) distribution, then the number generated would be the same.That's where set.seed() (or random.seed()) come in.
In a 'test' closure[1] the seed is made exogenous to the model, by switching on an equation for it that effectively switches off the assumed "inherent" randomness.
When the 'seed' equations are switched on, everything's back to being deterministic and normal unit-testing works.
[1] A closure is a partition of the variables in the model into 'given to the model' (exogenous) and 'determined within the model' (endogenous).
Models always has more variables (N) than equations (M > N), and so (M-N) values have to be given in order for the model to solve.
Selection of the (M-N) exogenous variables changes how the model 'works' (what things it is determining).
There are at least 5 different closures that are routine -
• testing* (outputs: depends);
• estimation (outputs: parameters and residuals);
• historical validation (outputs: 'naturally' endogenous variables);
• forecast (outputs: 'naturally' endogenous variables).
Historical validation closures are when you discover how bad a forecast your model would have produced if you ran the model in forecast mode, giving it exactly-correct values for the historical values used for estimation.
Testing has an asterisk because it is really a quasi-closure, because it adds equations that aren't really part of the model. It's only used to check that the code is correctly implemented, and has to happen before any estimation is done.Replies: @res
That is how it is supposed to work. The issue is that people are claiming setting the seed in the Imperial model does not produce deterministic results. From what I can see is there was a single case identified in issue 116 which showed non-determinism when running multi-threaded. The best discussion I have seen of this is between gavinpotter and weshinsley (I think he is a member of the development team) in this issue: https://github.com/mrc-ide/covid-sim/issues/165
IF weshinsley’s take is accurate then the non-determinism is probably not a big deal. BUT that raises the question of why don’t there appear to be regression tests. A comment from this review gives one idea:
https://www.armstrongeconomics.com/world-news/corruption/i-have-reviewed-fergusons-code/
“However, we haven’t had the time to work out a scalable and maintainable way of running the regression test in a way that allows a small amount of variation, but doesn’t let the figures drift over time.”
As for whether the seed is an input. It is an input to the code which is how I think of this. And BTW, if you are counting the seed as an exogenous variable then I think you need to give the model M-N+1 variables (including the seed as the 1) since it does not appear explicitly in the model equations.
Perhaps it would be more accurate to say the seed is both an input and an output. The starting seed is an input and the resulting seed after any given number of calls to the pseudo-random number generator have been made is an output. The starting seed is generally derived from some varying system value (e.g. time) if left unspecified.
Your use of “closure” seems nonstandard to me. I am used to definitions along these lines:
https://softwareengineering.stackexchange.com/questions/40454/what-is-a-closure
I mostly get what you mean from context (it’s similar to, but not quite the same as, the usage I am used to), but a reference that describes your usage would be helpful.
The decomposition into estimation, historical validation, and forecast is helpful. I am quite curious about what historical validation they have done. Especially given Ferguson’s poor prediction record in the past.
P.S. Your comment reads like you think I did not know about set.seed etc. Please don’t assume I am clueless. Also, your comment reads like it is tending towards complicated terminology for the sake of obfuscation rather than clarification. I consider that a bug, not a feature.
By 1982 it was common enough in quantitative economics that it was included in article titles, e.g.,
Rattso, J., 1982. Different macroclosures of the original Johansen model and their impact on policy evaluation, Journal of Policy Modeling March 1982, pp 85-97.
Even so, the same word was used (in a different sense) in the software world decades earlier, and this is kind of the intersection of mathematical modelling and software - so the confusion is understandable.It's certainly both from the point of view of the code, but it's actually neither from the point of view of the model. If the aim is to try to test if the code does the stuff implied by the model, seeds need to be stored to make certain that they're being used appropriately.
Sounds like splitting hairs, I know - "the code" and "the model" are pretty much considered to be interchangeable terms, but they're semantically different.
I use them interchangeably all the time, which is a bad habit I picked up from hanging around with GIS people - which is weird because they're terminological sticklers about 'projection' vs 'SRID'.
.
I certainly assumed you knew what set.seed did: the point was it's the answer to how to test code for a model that normally has 'fuzzy' stuff happening on purpose.
That's why the testing closure has an extra equation (in the code, but not the model) - to switch off the fuzz by setting a seed, which reduces the code to a deterministic system (the model is still inherently fuzzy).
As to ICL's code: the input seed won't be the only seed, I bet. There will be something somewhere that either ignores the set.seed equivalent, or some other undiscovered nonsense.
This is what happens when people hand down undocumented code over the course of two decades.Replies: @res
Sometimes the summary gets its own summary.
And I'm still not sure how much people actually read.Replies: @Chrisnonymous
Why would they read what you provide summaries for?
I’d be happy if they even read (and understood) the summary.
https://www.armstrongeconomics.com/world-news/corruption/i-have-reviewed-fergusons-code/
"However, we haven’t had the time to work out a scalable and maintainable way of running the regression test in a way that allows a small amount of variation, but doesn’t let the figures drift over time."As for whether the seed is an input. It is an input to the code which is how I think of this. And BTW, if you are counting the seed as an exogenous variable then I think you need to give the model M-N+1 variables (including the seed as the 1) since it does not appear explicitly in the model equations.Perhaps it would be more accurate to say the seed is both an input and an output. The starting seed is an input and the resulting seed after any given number of calls to the pseudo-random number generator have been made is an output. The starting seed is generally derived from some varying system value (e.g. time) if left unspecified.Your use of "closure" seems nonstandard to me. I am used to definitions along these lines:
https://softwareengineering.stackexchange.com/questions/40454/what-is-a-closure
I mostly get what you mean from context (it's similar to, but not quite the same as, the usage I am used to), but a reference that describes your usage would be helpful.The decomposition into estimation, historical validation, and forecast is helpful. I am quite curious about what historical validation they have done. Especially given Ferguson's poor prediction record in the past.P.S. Your comment reads like you think I did not know about set.seed etc. Please don't assume I am clueless. Also, your comment reads like it is tending towards complicated terminology for the sake of obfuscation rather than clarification. I consider that a bug, not a feature.Replies: @Kratoklastes
“Closure” used in the sense I’ve used it – the selection of model variables to set as inputs, so that the model is mathematically solvable – dates from the 1970s at the latest.
By 1982 it was common enough in quantitative economics that it was included in article titles, e.g.,
Rattso, J., 1982. Different macroclosures of the original Johansen model and their impact on policy evaluation, Journal of Policy Modeling March 1982, pp 85-97.
Even so, the same word was used (in a different sense) in the software world decades earlier, and this is kind of the intersection of mathematical modelling and software – so the confusion is understandable.
It’s certainly both from the point of view of the code, but it’s actually neither from the point of view of the model. If the aim is to try to test if the code does the stuff implied by the model, seeds need to be stored to make certain that they’re being used appropriately.
Sounds like splitting hairs, I know – “the code” and “the model” are pretty much considered to be interchangeable terms, but they’re semantically different.
I use them interchangeably all the time, which is a bad habit I picked up from hanging around with GIS people – which is weird because they’re terminological sticklers about ‘projection’ vs ‘SRID’.
.
I certainly assumed you knew what set.seed did: the point was it’s the answer to how to test code for a model that normally has ‘fuzzy’ stuff happening on purpose.
That’s why the testing closure has an extra equation (in the code, but not the model) – to switch off the fuzz by setting a seed, which reduces the code to a deterministic system (the model is still inherently fuzzy).
As to ICL’s code: the input seed won’t be the only seed, I bet. There will be something somewhere that either ignores the set.seed equivalent, or some other undiscovered nonsense.
This is what happens when people hand down undocumented code over the course of two decades.
The code/model distinction seems fairly similar to the code/algorithm distinction:
https://hackernoon.com/the-future-is-algorithms-not-code-64cacca3b908
But I'm not sure how exact that correspondence is. What do you think?Sounds likely to me as well.
By 1982 it was common enough in quantitative economics that it was included in article titles, e.g.,
Rattso, J., 1982. Different macroclosures of the original Johansen model and their impact on policy evaluation, Journal of Policy Modeling March 1982, pp 85-97.
Even so, the same word was used (in a different sense) in the software world decades earlier, and this is kind of the intersection of mathematical modelling and software - so the confusion is understandable.It's certainly both from the point of view of the code, but it's actually neither from the point of view of the model. If the aim is to try to test if the code does the stuff implied by the model, seeds need to be stored to make certain that they're being used appropriately.
Sounds like splitting hairs, I know - "the code" and "the model" are pretty much considered to be interchangeable terms, but they're semantically different.
I use them interchangeably all the time, which is a bad habit I picked up from hanging around with GIS people - which is weird because they're terminological sticklers about 'projection' vs 'SRID'.
.
I certainly assumed you knew what set.seed did: the point was it's the answer to how to test code for a model that normally has 'fuzzy' stuff happening on purpose.
That's why the testing closure has an extra equation (in the code, but not the model) - to switch off the fuzz by setting a seed, which reduces the code to a deterministic system (the model is still inherently fuzzy).
As to ICL's code: the input seed won't be the only seed, I bet. There will be something somewhere that either ignores the set.seed equivalent, or some other undiscovered nonsense.
This is what happens when people hand down undocumented code over the course of two decades.Replies: @res
Thanks for the clarifications. And introducing me to that use of “closure.”
I tend to do the same. But there are times when it is important to make the distinction. As here.
The code/model distinction seems fairly similar to the code/algorithm distinction:
https://hackernoon.com/the-future-is-algorithms-not-code-64cacca3b908
But I’m not sure how exact that correspondence is. What do you think?
Sounds likely to me as well.