|
Replies: 19
| visibility 1
|
All-TigerNet [12233]
TigerPulse: 100%
Posts: 5388
Joined: 9/12/04
|
|
|
|
 |
All-In [40875]
TigerPulse: 100%
Posts: 42932
Joined: 11/30/98
|
It was fairly obvious from day one that Ferguson was FOS
May 8, 2020, 7:58 AM
|
|
his predictions were crazy. The first thing he said was that China was lying and they had 100's of thousands of cases all the way in Feb.
"Based on this, Ferguson’s team calculates that, by 31 January, there were at least 24,000 new cases a day in Wuhan, which calls into question the current fall in case reports, which number around 3000 a day. This could also mean that total case numbers in China may now be as many as a million. If this is the case, and if all deaths in Wuhan are being detected, then, says the team, the overall death rate is only around 1 per cent – which matches Leung’s prediction."
this was back Feb 11
https://www.newscientist.com/article/2233269-how-bad-is-the-covid-19-coronavirus-outbreak-likely-to-get/
|
|
|
|
|
|
Orange Blooded [4545]
TigerPulse: 100%
Posts: 3357
Joined: 8/14/01
|
Re: It was fairly obvious from day one that Ferguson was FOS
May 8, 2020, 1:22 PM
|
|
The same guy who wrote the alarmist Imperial College report that said that 2M people in the US would die of COVID made similarly alarming projections in earlier pandemics and was way wrong those times.
His fellow epidemiologists don't think much of his models it seems...“In 2009, one of Ferguson’s models predicted 65,000 people could die from the Swine Flu outbreak in the UK — the final figure was below 500.” Business Insider also noted, “Michael Thrusfield, a professor of veterinary epidemiology at Edinburgh University, told the paper he had ‘déjà vu’ after reading the Imperial paper, saying Ferguson was responsible for excessive animal culling during the 2001 Foot and Mouth outbreak. Ferguson warned the government that 150,000 people could die. Six million animals were slaughtered as a precaution, costing the country billions in farming revenue. In the end, 200 people died.”
But hey, if he keeps making these projections, maybe he'll eventually be right, right? But it sure seems like his models are designed to get headlines than to be accurate. Maybe this is the one even. So far it doesn't look like it.
|
|
|
|
|
|
CU Guru [1786]
TigerPulse: 100%
Posts: 1111
Joined: 4/6/18
|
Re: Maybe this is why the US dumped the Imperial College model
May 8, 2020, 8:03 AM
|
|
Given the current climate of distrust for politicians and their cronies, this will not play well with about half the population. The other half seem to be the cronies.
I suppose my comment should be 'Why am I not surprised by this'.
|
|
|
|
|
|
Athletic Dir [881]
TigerPulse: 57%
Posts: 2125
Joined: 2/3/12
|
Very bad article
May 8, 2020, 8:12 AM
|
|
.
|
|
|
|
|
|
Orange Blooded [4563]
TigerPulse: 100%
Posts: 3141
Joined: 9/4/03
|
|
|
|
|
|
Athletic Dir [881]
TigerPulse: 57%
Posts: 2125
Joined: 2/3/12
|
That’s even worse
May 8, 2020, 8:43 AM
|
|
This is one tell:
“ It’s safe because the seed can be recorded and the same (pseudo-)random numbers produced from it in future. ”
That’s the opposite definition of a random number generator. The writer doesn’t understand how stochastic models work. Reading between the lines, when they fix all the input and model parameters, there is still some randomness with multiple cpus. If the average of the changes between runs is zero and the std is small, it shouldn’t matter. You run that model thousands or millions of times with random perturbations to the input and model parameters to generate an ensemble that you then use to get the upper and lower limits on the predictions.
Deterministic is when you change each parameter in a best-case and worst case way. If they did it that way, there would far more criticism from non-anonymous bloggers.
|
|
|
|
|
|
CU Medallion [73569]
TigerPulse: 100%
Posts: 78044
Joined: 11/30/98
|
lol, you didnt read very far but ok
May 8, 2020, 8:58 AM
|
|
But Edinburgh came back and reported that – even in single-threaded mode – they still see the problem. So Imperial’s understanding of the issue is wrong. Finally, Imperial admit there’s a bug by referencing a code change they’ve made that fixes it. The explanation given is “It looks like historically the second pair of seeds had been used at this point, to make the runs identical regardless of how the network was made, but that this had been changed when seed-resetting was implemented”. In other words, in the process of changing the model they made it non-replicable and never noticed.
|
|
|
|
|
|
Athletic Dir [881]
TigerPulse: 57%
Posts: 2125
Joined: 2/3/12
|
She got a lot wrong up front
May 8, 2020, 9:26 AM
|
|
I read all of it, but a lot of her complaints don’t matter. She’s probably a good coder if she worked for google, but she’s out of her depth here. If I’m running a monte carlo simulation and there’s a little more randomness than caused by the random function, I don’t care as long as the mean is zero and std is small. Every code has bugs. I always check a random function with a histogram (pdf) and frequency plot (psd), all using the same seed. It should produce a gaussian pdf and a flat spectrum.
The sources of overprediction by the imperial model are the assumptions, models, and model parameters, not a code bug.
|
|
|
|
|
|
110%er [9023]
TigerPulse: 100%
Posts: 7375
Joined: 4/9/11
|
Re: Read the code review
May 8, 2020, 10:56 AM
[ in reply to Read the code review ] |
|
Thanks for the website referral. Lots of information to go through!
|
|
|
|
|
|
110%er [6153]
TigerPulse: 100%
Posts: 4043
Joined: 2/15/11
|
Why is it a very bad article? Because it opines bad science***
May 8, 2020, 8:34 AM
[ in reply to Very bad article ] |
|
|
|
|
|
|
|
All-In [27366]
TigerPulse: 100%
Posts: 26233
Joined: 9/19/11
|
|
|
|
|
|
CU Medallion [64519]
TigerPulse: 100%
Posts: 88961
Joined: 3/27/01
|
I am reminded of the quote from Ghostbusters...
May 8, 2020, 8:35 AM
|
|
"We believe that the purpose of science is to serve mankind. You, however, seem to regard science as some kind of dodge... or hustle. Your theories are the worst kind of popular tripe, your methods are sloppy, and your conclusions are highly questionable! You are a poor scientist."
|
|
|
|
|
|
110%er [7157]
TigerPulse: 100%
Posts: 10133
Joined: 6/27/07
|
Everything was fine until the grid was shut down by @ickless
May 8, 2020, 10:08 AM
|
|
here - Ray Stanz
|
|
|
|
|
|
Legend [16472]
TigerPulse: 100%
Posts: 11652
Joined: 9/1/01
|
|
|
|
|
|
Oculus Spirit [78831]
TigerPulse: 100%
Posts: 26421
Joined: 12/6/98
|
I know a little bit about crappy code
May 8, 2020, 10:30 AM
|
|
not my own, of course
|
|
|
|
|
|
Legend [16472]
TigerPulse: 100%
Posts: 11652
Joined: 9/1/01
|
Must have been a missing punch card.........***
May 8, 2020, 11:20 AM
|
|
|
|
|
|
|
|
All-TigerNet [12270]
TigerPulse: 100%
Posts: 12594
Joined: 12/1/03
|
a scientist by definition:
May 8, 2020, 10:44 AM
|
|
"a person who passes as an expert on the basis of being able to turn out, with prolific fortitude, infinite quantities of incomprehensible formulae, calculated with micrometric precision from vague assumptions, which were based on debatable figures, taken from inconclusive experiments carried out with instruments of problematical accuracy by persons of doubtful reliability and questionable mentatlity."
|
|
|
|
|
|
CU Medallion [65037]
TigerPulse: 100%
Posts: 33173
Joined: 12/3/03
|
Looks like a Pile of Pink Poop.
May 8, 2020, 12:48 PM
|
|
|
|
|
|
|
|
Orange Blooded [4545]
TigerPulse: 100%
Posts: 3357
Joined: 8/14/01
|
Re: Maybe this is why the US dumped the Imperial College model
May 8, 2020, 3:01 PM
|
|
I spent some time looking at this code over lunch and have a few comments myself.
The code is labeled as C++, but is simply poorly written C being compiled by a C++ compiler. In fact, I'm pretty sure some of it was originally in fortran and has been converted to C using f2c (the tell-tales are there).
If anyone on my software development team had produced code like this, I would beat them with a stick. The code looks like something from people who've found a C reference manual from the 1980s and kept pounding on code until it only crashes sometimes instead of every time.
No coding standards from any reputable source (MISRA, etc.) are followed.
No code formatting standards seem to be in place.
A "most-telling" commit log comment exists on MANY files. It says "Squash history for public release" meaning that they erased the change logs for who knows how many bugs, changes, etc.? Cover-up much?
Even the most recent change logs notices seem to admit errors, e.g., "Lots of changes to make CT more realistic. Buggy"
An inordinate amount of code is used because clearly the authors do not understand how to use even the most basic standard library calls and wrote their own. E.g., several hundred lines to implement getopt() manually for each command-line parameter.
I would not trust the people who wrote this code to properly write software to compute body-mass-index, let alone a complex model like this.
And we stopped the world based on the output from this model.
I am highly reminded here of the similar bad state of affairs in the Mann climate change model software, which had many of the same sorts of problems.
I also have some exposure to other climate-study models. They all sucked, too.
The common thread here is that the researchers who do the math then try to write the code. It's not something they're trained to do, and they do it poorly. Epidemiologists should not write software any more than they should perform open-heart surgery.
|
|
|
|
|
|
Replies: 19
| visibility 1
|
|
|
50
