By now we have more than enough data to train a proper ML model with multiple variables that will spit out a decent range, yet we got 800-107k lives saved for the circuit breaker. Academics have not kept up with best practices in data modelling and now we're paying the price.
Data science (or any science) is about understanding properties, it is not about forecasting single or point outcomes.
If a distribution has a fat tail, most of the information is away from the centre of the distribution. For a fat tail, it is not possible or sensible to make a point forecast. All that is possible is a distributional forecast.
I don’t think Neil Ferguson's models are flawed because of poor data modelling practice. They have flaws, but the flaws are inevitable — wrong assumptions and high sensitivity are unavoidable in modelling the complex effects of something that has never happened before on the human population.
The sad truth is I’m afraid: point forecasts are close to useless and distributional forecasts are really hard.
I am much more of a COVID empiricist.
We can see what works by looking around the world. And we should put our faith in that, and not in modelling which does not set out to do what you want it to.
Ferguson has received a lot of criticism but much of it simply misunderstands what is even possible with this kind of modelling.
Trafalgar have just slapped a NC poll up on their site.
And you are never going to believe it, it has exactly the same demographics as the one from September.
Outstanding consistency
Likely they've got some techie who takes txt & robo push-button responses to push-polling questions (where the REAL point is message NOT the response) then re-configures results based on demographics obtained from source OTHER than the actual texts & calls.
Then Cahaly pumps it out as another Trafalgar Group "poll".
This is all supposition - but it DOES fit the known facts (including TG's own vague "explanation" re: methodology) AND is certainly right in the Lee Atwater - Karl Rove (remember him?) wheelhouse.
Nah, the polls appear g on their website are just made up. The 3 that appeared today and caused a stir are genuine-but-crap. Notice
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
Not really, it's been very disorganised and people have been proved to be irresponsible and not isolating after testing positive and the state has been left wanting in many areas. Any thoughts of exceptionalism should have been washed away by the virus response both from the state and the people.
The NHS, academic and pharmaceutical contribution is world class.
Lots of questions over the academic response over this, especially early on. Many so called experts not leveraging standard modern tools. Much that everybody laughed at the use of Excel for of the testing data, we saw with the likes of Ferguson model is was a similar shit show of coding / obselete programming language and paradigms.
We are still seeing it now with the model for how effective a circuit breaker might be providing nonsensical results.
The obsolete programming language was C, used for linux and android and therefore probably the most important and most commonly deployed language. (Obscure namedrop: I was once stuck in a lift with that there Dennis Ritchie.) I've not been following modelgate but gather some Microsofties had a go at porting Ferguson's code to C++ for no readily apparent reason except that is what programmers like to do: reimplement the wheel in a new language.
When all this is over, we might reflect that Dominic Cummings and pb's own @SeanT were right about one thing. Britain does need to invest more in research.
C is most definitely not an obsolete language it is still widely used for good reasons. That is however not to say its necessarily the best to use for those models. I can't comment on that because I haven't seen the models but C is I believe still one of the top ten languages and is still the best to choose often for embedded purposes. Like everything it is the right tool for the right purpose.
I would have more concern if the models were done in something live javascript than c.
John Carmack looked at the code and said it looked fine.
That is more than good enough for me.
Me too. The point I was making was merely that C is an old language but that doesn't make it obsolete. I wasn't arguing it was the right choice for this however I said I didn't know as hadn't seen the model.
Half the computer world still runs on C, apache web servers, linux, most embedded stuff etc.
It is effectively obselete for those supposed doing state of the art mathematical modelling, which was what i was talking about (not that servers run on loads of c code). That is what we were arguing, that they world leading experts are using totally out of date approaches.
It is like the using of excel to collate the testing data. It works, but nobody who knows what they are doing would use that.
I would take issue with you advocating as well for multithreading. It is a known source of errors due to not atomic updating of things due to programmer error.
If you want results fast go multithreaded and parallel. If you dont mind if it takes a few hours you are more likely to get reliable results with a single threaded app as you havent added possible error sources by hoping no one has forgotton to use a mutex/critical section/semaphore etc where they should have. Use it when you need to but be aware of the drawbacks
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
Not really, it's been very disorganised and people have been proved to be irresponsible and not isolating after testing positive and the state has been left wanting in many areas. Any thoughts of exceptionalism should have been washed away by the virus response both from the state and the people.
The NHS, academic and pharmaceutical contribution is world class.
Lots of questions over the academic response over this, especially early on. Many so called experts not leveraging standard modern tools. Much that everybody laughed at the use of Excel for of the testing data, we saw with the likes of Ferguson model is was a similar shit show of coding / obselete programming language and paradigms.
We are still seeing it now with the model for how effective a circuit breaker might be providing nonsensical results.
The obsolete programming language was C, used for linux and android and therefore probably the most important and most commonly deployed language. (Obscure namedrop: I was once stuck in a lift with that there Dennis Ritchie.) I've not been following modelgate but gather some Microsofties had a go at porting Ferguson's code to C++ for no readily apparent reason except that is what programmers like to do: reimplement the wheel in a new language.
When all this is over, we might reflect that Dominic Cummings and pb's own @SeanT were right about one thing. Britain does need to invest more in research.
Firstly, a portion of the model was auto-generated code from Fortran. Fortran...what did he code it on, a 486?
Secondly, nobody who knows what they are doing does statistical modelling in C these days...
Furthermore, it was single threaded, made no use of the likes of Intel MKL, Cuda or basically any modern programming suite of tools / libraries that people use for modelling complex problems.
You might want to check those python stats libraries. I'm told they are often just syntactic sugar over C and more often Fortran routines.
But you are right. No-one codes these models in C any more but Ferguson's model was a research tool written more than a decade ago, pressed back into service for want of anything better.
Yes and no...he is supposed to be a world leading expert in this field. His modelling was called on for swine flu, and it didn't work very well and since then he hasn't updated it, nor had a post-doc or even a PhD work on it.
That isn't how being a world leader is suppose to be. You should be constantly improving with the times.
The past 10 years has seen incredible progress in machine learning. He hasn't kept up with the times.
And this was the best we had. The original suggestion was academia were brilliant in response to this pandemic, it was just the government who were poorly prepared and crap.
Welcome back @LadyG - hope you managed your road trip safely.
Ta. I’m still on it. So far it’s been a mixture of highly enjoyable and very sobering. Saw lots of family, had quite jolly times, made the most of Cornish sun and good seafood. But today I met a friend in Devon who I haven’t seen for a year (partly because Covid). He gave me the full roster of troubles in one group of acquaintances: debt, despair, bankruptcy, divorce, death, the works.
And people are now obviously cracking up under the strain. God speed an effective vaccine.
Wait until the bills for this start to come in....everybody is going to be paying for this for the rest of their lives, and so are their kids.
What freaked me out today was a quote from a senior Swiss official (I forget where I read it, apologies) who warned his nation to expect a second wave “much much worse” than the first, including deaths.
Dunno why, but I always view the Swiss with a certain wary respect. Perhaps the most boringly sensible - or logically selfish - nation in the world.
If they really think the 2nd wave is going to be that bad - and early signs are grim - then EEEEESH
It does appear to be heading that way. Hopefully not. It could be brutal.
Does it? We’ve been entering this second wave for a month, now. Total UK critical cases, according to Worldometer data, is currently 743. Belgium - awash with new positive cases - about the same. Italy and Germany, interestingly behind us on new case numbers, have more in ICU, but not much above 1,000 each nationwide. So far, the second wave appears contagious but a lot less dangerous.
Yet we are damaging the economy with our policy response, as if it was the same.
ONS weekly deaths is doubling every 2 weeks.
Exponential is a killer. Nothing happens then everything happens at once.
Fine, but we are losing sight of the fact that the calibration is to protect the NHS from being overwhelmed, not to eliminate the virus, which - pending a vaccine - is impossible.
More people - and certainly more people who aren’t already very elderly - are dying of cancer and heart failure each week, and yet it would never enter our minds to incur this level of economic damage to drive down those deaths.
Welcome back @LadyG - hope you managed your road trip safely.
Ta. I’m still on it. So far it’s been a mixture of highly enjoyable and very sobering. Saw lots of family, had quite jolly times, made the most of Cornish sun and good seafood. But today I met a friend in Devon who I haven’t seen for a year (partly because Covid). He gave me the full roster of troubles in one group of acquaintances: debt, despair, bankruptcy, divorce, death, the works.
And people are now obviously cracking up under the strain. God speed an effective vaccine.
Wait until the bills for this start to come in....everybody is going to be paying for this for the rest of their lives, and so are their kids.
What freaked me out today was a quote from a senior Swiss official (I forget where I read it, apologies) who warned his nation to expect a second wave “much much worse” than the first, including deaths.
Dunno why, but I always view the Swiss with a certain wary respect. Perhaps the most boringly sensible - or logically selfish - nation in the world.
If they really think the 2nd wave is going to be that bad - and early signs are grim - then EEEEESH
It does appear to be heading that way. Hopefully not. It could be brutal.
Does it? We’ve been entering this second wave for a month, now. Total UK critical cases, according to Worldometer data, is currently 743. Belgium - awash with new positive cases - about the same. Italy and Germany, interestingly behind us on new case numbers, have more in ICU, but not much above 1,000 each nationwide. So far, the second wave appears contagious but a lot less dangerous.
Yet we are damaging the economy with our policy response, as if it was the same.
UK weekly deaths are doubling every 2 weeks.
Exponential is a killer. Nothing happens then everything happens at once.
Yes. Exactly. And deaths are a lagging indicator, of course
Basically, no one really knows. There are so many imponderables it is beyond the wit of man to know the future of this virus.
However the University of Washington Covid Model, which used to be laughably poor, now looks increasingly prescient, as they keep pumping in more data.
FWIW by Feb 2021 they predict 150,000 dead in the UK, and 385,000 dead in the USA. Grisly. Let’s hope they are wrong
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
For all the mistakes I do not think HMG have performed much worse than other governments to be fair
What we're seeing now is the difference between a Prime Minister who's on top of his brief (Cameron) and an extremely lazy and slapdash one (Johnson).
Cameron would have closed this down by Friday last week, and amended the narrative by pushing the issue onto his turf. He got criticised for being an "essay crisis" PM for that but at least he recognised a crisis when he saw one, and actually wrote the essay and got it in on time. He'd have learnt lessons from it on how to avoid it again.
Boris simply can't be arsed. He doesn't do his red boxes. He doesn't think even one step ahead and just lets events happen. Meanwhile various backbenches and civil servants either stick to the Government's original message or try and triangulate with the developing political arguments, risking them looking stupid and foolish as things develop (and ultimately) change.
Eventually when things spiral out of control Boris realises he has no choice but to backpedal. By that stage he's surrendered all initiative and looks like he's reacting to events rather than shaping them, because he is.
There's a reason Max Hastings, Michael Howard, David Cameron and all the mayoral deputies who used to work for him thought he was useless - because he is.
Asleep at the wheel - with a suspended drivers license AND an open bottle of gin in the glove box.
Isn't that a US crime, the open bottle, rather than an English one?
Last time I rented a car in UK, was told cup-holders were illegal. BUT cruising about with an open container of alcohol is AOK?
Why shouldn't it be? You are allowed to drink and drive, and in any case your passengers might like a drink. I've never heard that one about cup holders.
You are allowed to drink HOW MUCH and drive? As for cupholders, that's what I was told at time, can't remember by whom.
In US laws vary state to state, but open container inside vehicle is no-no everywhere; if you've got one with you, supposed to put it in truck (boot to you) or lock it in glove box (whatever you call it).
The limit is normally reckoned to be about a pint and a half of beer.
Which would have widely varying affects upon people of different weight & other relevant factors (whatever they are).
In US authorities almost always blood alcohol percents (determined by blood sample or more commonly by breathilizer) to determine if drivers are over the limit. Also field sobriety tests pretty common.
That's IF you get pulled over for erratic driving OR are involved in a accident and appear to be impaired.
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
For all the mistakes I do not think HMG have performed much worse than other governments to be fair
What we're seeing now is the difference between a Prime Minister who's on top of his brief (Cameron) and an extremely lazy and slapdash one (Johnson).
Cameron would have closed this down by Friday last week, and amended the narrative by pushing the issue onto his turf. He got criticised for being an "essay crisis" PM for that but at least he recognised a crisis when he saw one, and actually wrote the essay and got it in on time. He'd have learnt lessons from it on how to avoid it again.
Boris simply can't be arsed. He doesn't do his red boxes. He doesn't think even one step ahead and just lets events happen. Meanwhile various backbenches and civil servants either stick to the Government's original message or try and triangulate with the developing political arguments, risking them looking stupid and foolish as things develop (and ultimately) change.
Eventually when things spiral out of control Boris realises he has no choice but to backpedal. By that stage he's surrendered all initiative and looks like he's reacting to events rather than shaping them, because he is.
There's a reason Max Hastings, Michael Howard, David Cameron and all the mayoral deputies who used to work for him thought he was useless - because he is.
Asleep at the wheel - with a suspended drivers license AND an open bottle of gin in the glove box.
Isn't that a US crime, the open bottle, rather than an English one?
Last time I rented a car in UK, was told cup-holders were illegal. BUT cruising about with an open container of alcohol is AOK?
I'm not familiar with English as opposed to Scots law. But as far as I know it is fine. It is the blood content of alcohol that counts - though an open container would probably get you breathalysed if Mr Police observed it.
Believe US open container laws predate breath tests. Still think they are a good idea, as it discourages drivers from having a belt or two or more while actually driving.
Keep in mind that drunk driving is a VERY serious problem across the USA where even today car ownership (and culture) is a BIGGER deal than in Europe (including obscure offshore islets),
Welcome back @LadyG - hope you managed your road trip safely.
Ta. I’m still on it. So far it’s been a mixture of highly enjoyable and very sobering. Saw lots of family, had quite jolly times, made the most of Cornish sun and good seafood. But today I met a friend in Devon who I haven’t seen for a year (partly because Covid). He gave me the full roster of troubles in one group of acquaintances: debt, despair, bankruptcy, divorce, death, the works.
And people are now obviously cracking up under the strain. God speed an effective vaccine.
Wait until the bills for this start to come in....everybody is going to be paying for this for the rest of their lives, and so are their kids.
What freaked me out today was a quote from a senior Swiss official (I forget where I read it, apologies) who warned his nation to expect a second wave “much much worse” than the first, including deaths.
Dunno why, but I always view the Swiss with a certain wary respect. Perhaps the most boringly sensible - or logically selfish - nation in the world.
If they really think the 2nd wave is going to be that bad - and early signs are grim - then EEEEESH
It does appear to be heading that way. Hopefully not. It could be brutal.
Does it? We’ve been entering this second wave for a month, now. Total UK critical cases, according to Worldometer data, is currently 743. Belgium - awash with new positive cases - about the same. Italy and Germany, interestingly behind us on new case numbers, have more in ICU, but not much above 1,000 each nationwide. So far, the second wave appears contagious but a lot less dangerous.
Yet we are damaging the economy with our policy response, as if it was the same.
UK weekly deaths are doubling every 2 weeks.
Exponential is a killer. Nothing happens then everything happens at once.
Yes. Exactly. And deaths are a lagging indicator, of course
Basically, no one really knows. There are so many imponderables it is beyond the wit of man to know the future of this virus.
However the University of Washington Covid Model, which used to be laughably poor, now looks increasingly prescient, as they keep pumping in more data.
FWIW by Feb 2021 they predict 150,000 dead in the UK, and 385,000 dead in the USA. Grisly. Let’s hope they are wrong
The problem with the UoW model that caused us to take the piss was it was totally unable to even predict tomorrows death totals. At the time, it was literally todays reported number of deaths is 800, our model says...somewhere between 50 and 3000 will be reported tomorrow.
Trafalgar have just slapped a NC poll up on their site.
And you are never going to believe it, it has exactly the same demographics as the one from September.
Outstanding consistency
Likely they've got some techie who takes txt & robo push-button responses to push-polling questions (where the REAL point is message NOT the response) then re-configures results based on demographics obtained from source OTHER than the actual texts & calls.
Then Cahaly pumps it out as another Trafalgar Group "poll".
This is all supposition - but it DOES fit the known facts (including TG's own vague "explanation" re: methodology) AND is certainly right in the Lee Atwater - Karl Rove (remember him?) wheelhouse.
Nah, the polls appear g on their website are just made up. The 3 that appeared today and caused a stir are genuine-but-crap. Notice
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
Not really, it's been very disorganised and people have been proved to be irresponsible and not isolating after testing positive and the state has been left wanting in many areas. Any thoughts of exceptionalism should have been washed away by the virus response both from the state and the people.
The NHS, academic and pharmaceutical contribution is world class.
Lots of questions over the academic response over this, especially early on. Many so called experts not leveraging standard modern tools. Much that everybody laughed at the use of Excel for of the testing data, we saw with the likes of Ferguson model is was a similar shit show of coding / obselete programming language and paradigms.
We are still seeing it now with the model for how effective a circuit breaker might be providing nonsensical results.
The obsolete programming language was C, used for linux and android and therefore probably the most important and most commonly deployed language. (Obscure namedrop: I was once stuck in a lift with that there Dennis Ritchie.) I've not been following modelgate but gather some Microsofties had a go at porting Ferguson's code to C++ for no readily apparent reason except that is what programmers like to do: reimplement the wheel in a new language.
When all this is over, we might reflect that Dominic Cummings and pb's own @SeanT were right about one thing. Britain does need to invest more in research.
C is most definitely not an obsolete language it is still widely used for good reasons. That is however not to say its necessarily the best to use for those models. I can't comment on that because I haven't seen the models but C is I believe still one of the top ten languages and is still the best to choose often for embedded purposes. Like everything it is the right tool for the right purpose.
I would have more concern if the models were done in something live javascript than c.
John Carmack looked at the code and said it looked fine.
That is more than good enough for me.
Me too. The point I was making was merely that C is an old language but that doesn't make it obsolete. I wasn't arguing it was the right choice for this however I said I didn't know as hadn't seen the model.
Half the computer world still runs on C, apache web servers, linux, most embedded stuff etc.
Half of the computing world isn't trying to predict a viral replication model within a modelled real world scenario. It is what ML is made to do, using Python which has a whole host of ML libraries and can use GPU acceleration would be the correct way of doing it. I couldn't imagine anyone in my industry creating a predictive model in C and banking is at the forefront of ML, C is just old fashioned and unsuitable for using Tensorflow or PyTorch either of which are extensively used for deep learning which is what a problem like this needs.
Incidentally I am in a very pleasant hotel in Bristol - indeed, The Bristol Hotel.
It is highly covid cautious in its check-in procedures, elevator protocols, room cleaning, etc.
Yet the restaurant is absolutely packed and rocking with maskless people quaffing wine and expelling aerosols all over the shop. There are lots of bars and pizzerias and sushi bars all around, and they are similarly busy. It’s very upbeat.... but I just don’t see how we avoid following Italy, France, Spain, etc, into a pretty severe second lockdown and early curfews. Britain is not exceptionally immune. Just late to the party. Again.
On the upside Bristol - a city I barely knew - turns out to be highly beguiling. A hidden gem. When the Wuhan Flu is over I might return to buy a plague-vacated Georgian house for £10.
Welcome back @LadyG - hope you managed your road trip safely.
Ta. I’m still on it. So far it’s been a mixture of highly enjoyable and very sobering. Saw lots of family, had quite jolly times, made the most of Cornish sun and good seafood. But today I met a friend in Devon who I haven’t seen for a year (partly because Covid). He gave me the full roster of troubles in one group of acquaintances: debt, despair, bankruptcy, divorce, death, the works.
And people are now obviously cracking up under the strain. God speed an effective vaccine.
Wait until the bills for this start to come in....everybody is going to be paying for this for the rest of their lives, and so are their kids.
What freaked me out today was a quote from a senior Swiss official (I forget where I read it, apologies) who warned his nation to expect a second wave “much much worse” than the first, including deaths.
Dunno why, but I always view the Swiss with a certain wary respect. Perhaps the most boringly sensible - or logically selfish - nation in the world.
If they really think the 2nd wave is going to be that bad - and early signs are grim - then EEEEESH
It does appear to be heading that way. Hopefully not. It could be brutal.
Does it? We’ve been entering this second wave for a month, now. Total UK critical cases, according to Worldometer data, is currently 743. Belgium - awash with new positive cases - about the same. Italy and Germany, interestingly behind us on new case numbers, have more in ICU, but not much above 1,000 each nationwide. So far, the second wave appears contagious but a lot less dangerous.
Yet we are damaging the economy with our policy response, as if it was the same.
ONS weekly deaths is doubling every 2 weeks.
Exponential is a killer. Nothing happens then everything happens at once.
Fine, but we are losing sight of the fact that the calibration is to protect the NHS from being overwhelmed, not to eliminate the virus, which - pending a vaccine - is impossible.
More people - and certainly more people who aren’t already very elderly - are dying of cancer and heart failure each week, and yet it would never enter our minds to incur this level of economic damage to drive down those deaths.
A thousand people are going to die of Covid this week. So next month 4000 people will die of Covid a week.
To get to those level of figures the hospitals will be flooded with patients.
Trafalgar have just slapped a NC poll up on their site.
And you are never going to believe it, it has exactly the same demographics as the one from September.
Outstanding consistency
Likely they've got some techie who takes txt & robo push-button responses to push-polling questions (where the REAL point is message NOT the response) then re-configures results based on demographics obtained from source OTHER than the actual texts & calls.
Then Cahaly pumps it out as another Trafalgar Group "poll".
This is all supposition - but it DOES fit the known facts (including TG's own vague "explanation" re: methodology) AND is certainly right in the Lee Atwater - Karl Rove (remember him?) wheelhouse.
Nah, the polls appear g on their website are just made up. The 3 that appeared today and caused a stir are genuine-but-crap. Notice
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
Not really, it's been very disorganised and people have been proved to be irresponsible and not isolating after testing positive and the state has been left wanting in many areas. Any thoughts of exceptionalism should have been washed away by the virus response both from the state and the people.
The NHS, academic and pharmaceutical contribution is world class.
Lots of questions over the academic response over this, especially early on. Many so called experts not leveraging standard modern tools. Much that everybody laughed at the use of Excel for of the testing data, we saw with the likes of Ferguson model is was a similar shit show of coding / obselete programming language and paradigms.
We are still seeing it now with the model for how effective a circuit breaker might be providing nonsensical results.
The obsolete programming language was C, used for linux and android and therefore probably the most important and most commonly deployed language. (Obscure namedrop: I was once stuck in a lift with that there Dennis Ritchie.) I've not been following modelgate but gather some Microsofties had a go at porting Ferguson's code to C++ for no readily apparent reason except that is what programmers like to do: reimplement the wheel in a new language.
When all this is over, we might reflect that Dominic Cummings and pb's own @SeanT were right about one thing. Britain does need to invest more in research.
C is most definitely not an obsolete language it is still widely used for good reasons. That is however not to say its necessarily the best to use for those models. I can't comment on that because I haven't seen the models but C is I believe still one of the top ten languages and is still the best to choose often for embedded purposes. Like everything it is the right tool for the right purpose.
I would have more concern if the models were done in something live javascript than c.
John Carmack looked at the code and said it looked fine.
That is more than good enough for me.
Carmack is a genius and an incredible programmer. He is, however, a game developer and probably dreams in C++. I know I did when I was in that industry.
I would take issue with you advocating as well for multithreading. It is a known source of errors due to not atomic updating of things due to programmer error.
If you want results fast go multithreaded and parallel. If you dont mind if it takes a few hours you are more likely to get reliable results with a single threaded app as you havent added possible error sources by hoping no one has forgotton to use a mutex/critical section/semaphore etc where they should have. Use it when you need to but be aware of the drawbacks
Really the issue is not the language, or performance. My phone would run rings round virtually any supercomputer being used for scientific modelling before the 1990s. It's the approach that is antiquated. It would be anitquated even if programmed in the latest fashionable language with all the bells and whistles employed.
For quite a while one of the best COVID-19 models was one guy from MIT applying machine learning to an off-the-shelf model. He was consitently beating all sorts of fancy national labs and universities, many of which were so crap at modelling that they could not even beat the baseline of repeating the previous measured change. To be fair modelling has improved dramatically over the last six months or so, but I don't think it's unreasonable to state that the bulk of epidemiological modelling when the pandemic hit was a load of rubbish.
Trafalgar have just slapped a NC poll up on their site.
And you are never going to believe it, it has exactly the same demographics as the one from September.
Outstanding consistency
Likely they've got some techie who takes txt & robo push-button responses to push-polling questions (where the REAL point is message NOT the response) then re-configures results based on demographics obtained from source OTHER than the actual texts & calls.
Then Cahaly pumps it out as another Trafalgar Group "poll".
This is all supposition - but it DOES fit the known facts (including TG's own vague "explanation" re: methodology) AND is certainly right in the Lee Atwater - Karl Rove (remember him?) wheelhouse.
Nah, the polls appear g on their website are just made up. The 3 that appeared today and caused a stir are genuine-but-crap. Notice
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
Not really, it's been very disorganised and people have been proved to be irresponsible and not isolating after testing positive and the state has been left wanting in many areas. Any thoughts of exceptionalism should have been washed away by the virus response both from the state and the people.
The NHS, academic and pharmaceutical contribution is world class.
Lots of questions over the academic response over this, especially early on. Many so called experts not leveraging standard modern tools. Much that everybody laughed at the use of Excel for of the testing data, we saw with the likes of Ferguson model is was a similar shit show of coding / obselete programming language and paradigms.
We are still seeing it now with the model for how effective a circuit breaker might be providing nonsensical results.
The obsolete programming language was C, used for linux and android and therefore probably the most important and most commonly deployed language. (Obscure namedrop: I was once stuck in a lift with that there Dennis Ritchie.) I've not been following modelgate but gather some Microsofties had a go at porting Ferguson's code to C++ for no readily apparent reason except that is what programmers like to do: reimplement the wheel in a new language.
When all this is over, we might reflect that Dominic Cummings and pb's own @SeanT were right about one thing. Britain does need to invest more in research.
C is most definitely not an obsolete language it is still widely used for good reasons. That is however not to say its necessarily the best to use for those models. I can't comment on that because I haven't seen the models but C is I believe still one of the top ten languages and is still the best to choose often for embedded purposes. Like everything it is the right tool for the right purpose.
I would have more concern if the models were done in something live javascript than c.
John Carmack looked at the code and said it looked fine.
That is more than good enough for me.
Me too. The point I was making was merely that C is an old language but that doesn't make it obsolete. I wasn't arguing it was the right choice for this however I said I didn't know as hadn't seen the model.
Half the computer world still runs on C, apache web servers, linux, most embedded stuff etc.
Half of the computing world isn't trying to predict a viral replication model within a modelled real world scenario. It is what ML is made to do, using Python which has a whole host of ML libraries and can use GPU acceleration would be the correct way of doing it. I couldn't imagine anyone in my industry creating a predictive model in C and banking is at the forefront of ML, C is just old fashioned and unsuitable for using Tensorflow or PyTorch either of which are extensively used for deep learning which is what a problem like this needs.
Sigh does no one read what you write here anymore. I didn't argue c was the correct tool I merely asserted it wasn't an obsolete language. I said the right tool for the right job. I suspect in the case of covid however the model itself is the flaw rather than the language it is written in.
Further as to machine learning personally that is over hyped and unproven. We have a huge amount of examples of machine learning where what the machine learnt isn't what people thought they had taught it. Facial recognition, predictive crime etc being the two most egregious but lots of others out there.
Welcome back @LadyG - hope you managed your road trip safely.
Ta. I’m still on it. So far it’s been a mixture of highly enjoyable and very sobering. Saw lots of family, had quite jolly times, made the most of Cornish sun and good seafood. But today I met a friend in Devon who I haven’t seen for a year (partly because Covid). He gave me the full roster of troubles in one group of acquaintances: debt, despair, bankruptcy, divorce, death, the works.
And people are now obviously cracking up under the strain. God speed an effective vaccine.
Wait until the bills for this start to come in....everybody is going to be paying for this for the rest of their lives, and so are their kids.
What freaked me out today was a quote from a senior Swiss official (I forget where I read it, apologies) who warned his nation to expect a second wave “much much worse” than the first, including deaths.
Dunno why, but I always view the Swiss with a certain wary respect. Perhaps the most boringly sensible - or logically selfish - nation in the world.
If they really think the 2nd wave is going to be that bad - and early signs are grim - then EEEEESH
It does appear to be heading that way. Hopefully not. It could be brutal.
Does it? We’ve been entering this second wave for a month, now. Total UK critical cases, according to Worldometer data, is currently 743. Belgium - awash with new positive cases - about the same. Italy and Germany, interestingly behind us on new case numbers, have more in ICU, but not much above 1,000 each nationwide. So far, the second wave appears contagious but a lot less dangerous.
Yet we are damaging the economy with our policy response, as if it was the same.
UK weekly deaths are doubling every 2 weeks.
Exponential is a killer. Nothing happens then everything happens at once.
Yes. Exactly. And deaths are a lagging indicator, of course
Basically, no one really knows. There are so many imponderables it is beyond the wit of man to know the future of this virus.
However the University of Washington Covid Model, which used to be laughably poor, now looks increasingly prescient, as they keep pumping in more data.
FWIW by Feb 2021 they predict 150,000 dead in the UK, and 385,000 dead in the USA. Grisly. Let’s hope they are wrong
The problem with the UoW model that caused us to take the piss was it was totally unable to even predict tomorrows death totals. At the time, it was literally todays reported number of deaths is 800, our model says...somewhere between 50 and 3000 will be reported tomorrow.
Yes, I remember it well. It was kind of entertaining, how ludicrously it wobbled about.
But quietly and relentlessly they have been honing it, and now it appears rather on point.
385,000 dead in the USA by Feb 1 2021 is actually optimistic compared to some other forecasts.
Trafalgar have just slapped a NC poll up on their site.
And you are never going to believe it, it has exactly the same demographics as the one from September.
Outstanding consistency
Likely they've got some techie who takes txt & robo push-button responses to push-polling questions (where the REAL point is message NOT the response) then re-configures results based on demographics obtained from source OTHER than the actual texts & calls.
Then Cahaly pumps it out as another Trafalgar Group "poll".
This is all supposition - but it DOES fit the known facts (including TG's own vague "explanation" re: methodology) AND is certainly right in the Lee Atwater - Karl Rove (remember him?) wheelhouse.
Nah, the polls appear g on their website are just made up. The 3 that appeared today and caused a stir are genuine-but-crap. Notice
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
Not really, it's been very disorganised and people have been proved to be irresponsible and not isolating after testing positive and the state has been left wanting in many areas. Any thoughts of exceptionalism should have been washed away by the virus response both from the state and the people.
The NHS, academic and pharmaceutical contribution is world class.
Lots of questions over the academic response over this, especially early on. Many so called experts not leveraging standard modern tools. Much that everybody laughed at the use of Excel for of the testing data, we saw with the likes of Ferguson model is was a similar shit show of coding / obselete programming language and paradigms.
We are still seeing it now with the model for how effective a circuit breaker might be providing nonsensical results.
The obsolete programming language was C, used for linux and android and therefore probably the most important and most commonly deployed language. (Obscure namedrop: I was once stuck in a lift with that there Dennis Ritchie.) I've not been following modelgate but gather some Microsofties had a go at porting Ferguson's code to C++ for no readily apparent reason except that is what programmers like to do: reimplement the wheel in a new language.
When all this is over, we might reflect that Dominic Cummings and pb's own @SeanT were right about one thing. Britain does need to invest more in research.
C is most definitely not an obsolete language it is still widely used for good reasons. That is however not to say its necessarily the best to use for those models. I can't comment on that because I haven't seen the models but C is I believe still one of the top ten languages and is still the best to choose often for embedded purposes. Like everything it is the right tool for the right purpose.
I would have more concern if the models were done in something live javascript than c.
John Carmack looked at the code and said it looked fine.
That is more than good enough for me.
Me too. The point I was making was merely that C is an old language but that doesn't make it obsolete. I wasn't arguing it was the right choice for this however I said I didn't know as hadn't seen the model.
Half the computer world still runs on C, apache web servers, linux, most embedded stuff etc.
Half of the computing world isn't trying to predict a viral replication model within a modelled real world scenario. It is what ML is made to do, using Python which has a whole host of ML libraries and can use GPU acceleration would be the correct way of doing it. I couldn't imagine anyone in my industry creating a predictive model in C and banking is at the forefront of ML, C is just old fashioned and unsuitable for using Tensorflow or PyTorch either of which are extensively used for deep learning which is what a problem like this needs.
This of course isn't even an academia vs industry thing. Every decent computer scientist who does research in ML or mathematical modelling is basing their work around this modern suite of tools.
Furthermore it isnt just a C vs python thing. The world of mathematical modelling has been revolutionized in the past 10 years (and not just deep neural nets).
Instead the world leading virus modeller is still using a model from 10+ years ago.
Trafalgar have just slapped a NC poll up on their site.
And you are never going to believe it, it has exactly the same demographics as the one from September.
Outstanding consistency
Likely they've got some techie who takes txt & robo push-button responses to push-polling questions (where the REAL point is message NOT the response) then re-configures results based on demographics obtained from source OTHER than the actual texts & calls.
Then Cahaly pumps it out as another Trafalgar Group "poll".
This is all supposition - but it DOES fit the known facts (including TG's own vague "explanation" re: methodology) AND is certainly right in the Lee Atwater - Karl Rove (remember him?) wheelhouse.
Nah, the polls appear g on their website are just made up. The 3 that appeared today and caused a stir are genuine-but-crap. Notice
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
Not really, it's been very disorganised and people have been proved to be irresponsible and not isolating after testing positive and the state has been left wanting in many areas. Any thoughts of exceptionalism should have been washed away by the virus response both from the state and the people.
The NHS, academic and pharmaceutical contribution is world class.
Lots of questions over the academic response over this, especially early on. Many so called experts not leveraging standard modern tools. Much that everybody laughed at the use of Excel for of the testing data, we saw with the likes of Ferguson model is was a similar shit show of coding / obselete programming language and paradigms.
We are still seeing it now with the model for how effective a circuit breaker might be providing nonsensical results.
The obsolete programming language was C, used for linux and android and therefore probably the most important and most commonly deployed language. (Obscure namedrop: I was once stuck in a lift with that there Dennis Ritchie.) I've not been following modelgate but gather some Microsofties had a go at porting Ferguson's code to C++ for no readily apparent reason except that is what programmers like to do: reimplement the wheel in a new language.
When all this is over, we might reflect that Dominic Cummings and pb's own @SeanT were right about one thing. Britain does need to invest more in research.
C is most definitely not an obsolete language it is still widely used for good reasons. That is however not to say its necessarily the best to use for those models. I can't comment on that because I haven't seen the models but C is I believe still one of the top ten languages and is still the best to choose often for embedded purposes. Like everything it is the right tool for the right purpose.
I would have more concern if the models were done in something live javascript than c.
John Carmack looked at the code and said it looked fine.
That is more than good enough for me.
Me too. The point I was making was merely that C is an old language but that doesn't make it obsolete. I wasn't arguing it was the right choice for this however I said I didn't know as hadn't seen the model.
Half the computer world still runs on C, apache web servers, linux, most embedded stuff etc.
Half of the computing world isn't trying to predict a viral replication model within a modelled real world scenario. It is what ML is made to do, using Python which has a whole host of ML libraries and can use GPU acceleration would be the correct way of doing it. I couldn't imagine anyone in my industry creating a predictive model in C and banking is at the forefront of ML, C is just old fashioned and unsuitable for using Tensorflow or PyTorch either of which are extensively used for deep learning which is what a problem like this needs.
I would probably be a little less strident than you about this. Yes, my expectation would be someone using Python, R, or Scala. But domain knowledge is at least as important as using the most up-to-date tools. I would give the task to a subject matter expert using older tools long before I invited someone to model it using the latest tools but who only knew about Netflix recommendations and SEO optimisation from their work experience. Of course, if would be best to have both.
For quite a while one of the best COVID-19 models was one guy from MIT applying machine learning to an off-the-shelf model. He was consitently beating all sorts of fancy national labs and universities, many of which were so crap at modelling that they could not even beat the baseline of repeating the previous measured change. To be fair modelling has improved dramatically over the last six months or so, but I don't think it's unreasonable to state that the bulk of epidemiological modelling when the pandemic hit was a load of rubbish.
Can we have a link please to the work of this guy from MIT ? Thanks.
Trafalgar have just slapped a NC poll up on their site.
And you are never going to believe it, it has exactly the same demographics as the one from September.
Outstanding consistency
Likely they've got some techie who takes txt & robo push-button responses to push-polling questions (where the REAL point is message NOT the response) then re-configures results based on demographics obtained from source OTHER than the actual texts & calls.
Then Cahaly pumps it out as another Trafalgar Group "poll".
This is all supposition - but it DOES fit the known facts (including TG's own vague "explanation" re: methodology) AND is certainly right in the Lee Atwater - Karl Rove (remember him?) wheelhouse.
Nah, the polls appear g on their website are just made up. The 3 that appeared today and caused a stir are genuine-but-crap. Notice
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
Not really, it's been very disorganised and people have been proved to be irresponsible and not isolating after testing positive and the state has been left wanting in many areas. Any thoughts of exceptionalism should have been washed away by the virus response both from the state and the people.
The NHS, academic and pharmaceutical contribution is world class.
Lots of questions over the academic response over this, especially early on. Many so called experts not leveraging standard modern tools. Much that everybody laughed at the use of Excel for of the testing data, we saw with the likes of Ferguson model is was a similar shit show of coding / obselete programming language and paradigms.
We are still seeing it now with the model for how effective a circuit breaker might be providing nonsensical results.
The obsolete programming language was C, used for linux and android and therefore probably the most important and most commonly deployed language. (Obscure namedrop: I was once stuck in a lift with that there Dennis Ritchie.) I've not been following modelgate but gather some Microsofties had a go at porting Ferguson's code to C++ for no readily apparent reason except that is what programmers like to do: reimplement the wheel in a new language.
When all this is over, we might reflect that Dominic Cummings and pb's own @SeanT were right about one thing. Britain does need to invest more in research.
C is most definitely not an obsolete language it is still widely used for good reasons. That is however not to say its necessarily the best to use for those models. I can't comment on that because I haven't seen the models but C is I believe still one of the top ten languages and is still the best to choose often for embedded purposes. Like everything it is the right tool for the right purpose.
I would have more concern if the models were done in something live javascript than c.
John Carmack looked at the code and said it looked fine.
That is more than good enough for me.
Me too. The point I was making was merely that C is an old language but that doesn't make it obsolete. I wasn't arguing it was the right choice for this however I said I didn't know as hadn't seen the model.
Half the computer world still runs on C, apache web servers, linux, most embedded stuff etc.
Half of the computing world isn't trying to predict a viral replication model within a modelled real world scenario. It is what ML is made to do, using Python which has a whole host of ML libraries and can use GPU acceleration would be the correct way of doing it. I couldn't imagine anyone in my industry creating a predictive model in C and banking is at the forefront of ML, C is just old fashioned and unsuitable for using Tensorflow or PyTorch either of which are extensively used for deep learning which is what a problem like this needs.
Sigh does no one read what you write here anymore. I didn't argue c was the correct tool I merely asserted it wasn't an obsolete language. I said the right tool for the right job. I suspect in the case of covid however the model itself is the flaw rather than the language it is written in.
Further as to machine learning personally that is over hyped and unproven. We have a huge amount of examples of machine learning where what the machine learnt isn't what people thought they had taught it. Facial recognition, predictive crime etc being the two most egregious but lots of others out there.
I think the point being made is that him using C shows how outdated his approach is.
ML is great for these kinds of problems because people are predictably unpredictable. Predicting a person's behaviour is impossible, predicting population behaviour isn't and it's one of the major uses of it.
Agree that facial recognition and crime models are a pile of dogshit, but that doesn't invalidate the approach to other more suitable problems. Viral replication is probably one of the most suitable problems for ML as it's a multivariate analysis and we have a huge amount of data available now. Another good area is medical outcome modelling, a friend of mine works for Babylon and they've been doing some incredible work on it.
Incidentally I am in a very pleasant hotel in Bristol - indeed, The Bristol Hotel.
It is highly covid cautious in its check-in procedures, elevator protocols, room cleaning, etc.
Yet the restaurant is absolutely packed and rocking with maskless people quaffing wine and expelling aerosols all over the shop. There are lots of bars and pizzerias and sushi bars all around, and they are similarly busy. It’s very upbeat.... but I just don’t see how we avoid following Italy, France, Spain, etc, into a pretty severe second lockdown and early curfews. Britain is not exceptionally immune. Just late to the party. Again.
On the upside Bristol - a city I barely knew - turns out to be highly beguiling. A hidden gem. When the Wuhan Flu is over I might return to buy a plague-vacated Georgian house for £10.
Hope everything is shipshape and Bristol fashion!
Bristol was the favorite British city of an old friend of mine. She only visited it once but had a really good time and really loved the place. Which was HUGELY important to early British colonialism and mercantilism, and to the settlement and development of (what became) the United States.
Both lifeboats are returning to base but the tanker is still sailing in circles out to sea. Doesn’t look like we are in any hurry to bring it to port. Presumably the stowaways have all been lifted off.
Incidentally I am in a very pleasant hotel in Bristol - indeed, The Bristol Hotel.
It is highly covid cautious in its check-in procedures, elevator protocols, room cleaning, etc.
Yet the restaurant is absolutely packed and rocking with maskless people quaffing wine and expelling aerosols all over the shop. There are lots of bars and pizzerias and sushi bars all around, and they are similarly busy. It’s very upbeat.... but I just don’t see how we avoid following Italy, France, Spain, etc, into a pretty severe second lockdown and early curfews. Britain is not exceptionally immune. Just late to the party. Again.
On the upside Bristol - a city I barely knew - turns out to be highly beguiling. A hidden gem. When the Wuhan Flu is over I might return to buy a plague-vacated Georgian house for £10.
More likely £10m once we discover the limits of QE and remember what hyperinflation is about.
Trafalgar have just slapped a NC poll up on their site.
And you are never going to believe it, it has exactly the same demographics as the one from September.
Outstanding consistency
Likely they've got some techie who takes txt & robo push-button responses to push-polling questions (where the REAL point is message NOT the response) then re-configures results based on demographics obtained from source OTHER than the actual texts & calls.
Then Cahaly pumps it out as another Trafalgar Group "poll".
This is all supposition - but it DOES fit the known facts (including TG's own vague "explanation" re: methodology) AND is certainly right in the Lee Atwater - Karl Rove (remember him?) wheelhouse.
Nah, the polls appear g on their website are just made up. The 3 that appeared today and caused a stir are genuine-but-crap. Notice
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
Not really, it's been very disorganised and people have been proved to be irresponsible and not isolating after testing positive and the state has been left wanting in many areas. Any thoughts of exceptionalism should have been washed away by the virus response both from the state and the people.
The NHS, academic and pharmaceutical contribution is world class.
Lots of questions over the academic response over this, especially early on. Many so called experts not leveraging standard modern tools. Much that everybody laughed at the use of Excel for of the testing data, we saw with the likes of Ferguson model is was a similar shit show of coding / obselete programming language and paradigms.
We are still seeing it now with the model for how effective a circuit breaker might be providing nonsensical results.
The obsolete programming language was C, used for linux and android and therefore probably the most important and most commonly deployed language. (Obscure namedrop: I was once stuck in a lift with that there Dennis Ritchie.) I've not been following modelgate but gather some Microsofties had a go at porting Ferguson's code to C++ for no readily apparent reason except that is what programmers like to do: reimplement the wheel in a new language.
When all this is over, we might reflect that Dominic Cummings and pb's own @SeanT were right about one thing. Britain does need to invest more in research.
C is most definitely not an obsolete language it is still widely used for good reasons. That is however not to say its necessarily the best to use for those models. I can't comment on that because I haven't seen the models but C is I believe still one of the top ten languages and is still the best to choose often for embedded purposes. Like everything it is the right tool for the right purpose.
I would have more concern if the models were done in something live javascript than c.
John Carmack looked at the code and said it looked fine.
That is more than good enough for me.
Me too. The point I was making was merely that C is an old language but that doesn't make it obsolete. I wasn't arguing it was the right choice for this however I said I didn't know as hadn't seen the model.
Half the computer world still runs on C, apache web servers, linux, most embedded stuff etc.
Half of the computing world isn't trying to predict a viral replication model within a modelled real world scenario. It is what ML is made to do, using Python which has a whole host of ML libraries and can use GPU acceleration would be the correct way of doing it. I couldn't imagine anyone in my industry creating a predictive model in C and banking is at the forefront of ML, C is just old fashioned and unsuitable for using Tensorflow or PyTorch either of which are extensively used for deep learning which is what a problem like this needs.
Sigh does no one read what you write here anymore. I didn't argue c was the correct tool I merely asserted it wasn't an obsolete language. I said the right tool for the right job. I suspect in the case of covid however the model itself is the flaw rather than the language it is written in.
Further as to machine learning personally that is over hyped and unproven. We have a huge amount of examples of machine learning where what the machine learnt isn't what people thought they had taught it. Facial recognition, predictive crime etc being the two most egregious but lots of others out there.
I think the point being made is that him using C shows how outdated his approach is.
ML is great for these kinds of problems because people are predictably unpredictable. Predicting a person's behaviour is impossible, predicting population behaviour isn't and it's one of the major uses of it.
Agree that facial recognition and crime models are a pile of dogshit, but that doesn't invalidate the approach to other more suitable problems. Viral replication is probably one of the most suitable problems for ML as it's a multivariate analysis and we have a huge amount of data available now. Another good area is medical outcome modelling, a friend of mine works for Babylon and they've been doing some incredible work on it.
Don't mistake me I am not saying machine learning doesnt have a place. Sadly like much in the it world it almost feels like its becoming the silver bullet.....whatever the problem machine learning is the answer or block chain
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
Not really, it's been very disorganised and people have been proved to be irresponsible and not isolating after testing positive and the state has been left wanting in many areas. Any thoughts of exceptionalism should have been washed away by the virus response both from the state and the people.
The NHS, academic and pharmaceutical contribution is world class.
Lots of questions over the academic response over this, especially early on. Many so called experts not leveraging standard modern tools. Much that everybody laughed at the use of Excel for of the testing data, we saw with the likes of Ferguson model is was a similar shit show of coding / obselete programming language and paradigms.
We are still seeing it now with the model for how effective a circuit breaker might be providing nonsensical results.
The obsolete programming language was C, used for linux and android and therefore probably the most important and most commonly deployed language. (Obscure namedrop: I was once stuck in a lift with that there Dennis Ritchie.) I've not been following modelgate but gather some Microsofties had a go at porting Ferguson's code to C++ for no readily apparent reason except that is what programmers like to do: reimplement the wheel in a new language.
When all this is over, we might reflect that Dominic Cummings and pb's own @SeanT were right about one thing. Britain does need to invest more in research.
Firstly, a portion of the model was auto-generated code from Fortran. Fortran...what did he code it on, a 486?
Secondly, nobody who knows what they are doing does statistical modelling in C these days...
Furthermore, it was single threaded, made no use of the likes of Intel MKL, Cuda or basically any modern programming suite of tools / libraries that people use for modelling complex problems.
Given how simple the model is fundamentally, going parallel would have been pointless.
My point is today anybody who knows what they are doing doesn't have to make that as a big design decision. Even if you aren't a coding whizz, it is trivial to make multi-threaded, use GPU, etc from the get go and the use of something like MKL (or a library built on it) is standard if you are considering doing mathematical modelling.
From talking to some people who were peripherally involved in reimplementing the model (multiple different groups, multiple languages) - it runs fine single threaded.
The interesting bit would be getting the ML "thinkers" to have a look at the whole area - looks like that wasn't in Ferguson's wheelhouse.
Welcome back @LadyG - hope you managed your road trip safely.
Ta. I’m still on it. So far it’s been a mixture of highly enjoyable and very sobering. Saw lots of family, had quite jolly times, made the most of Cornish sun and good seafood. But today I met a friend in Devon who I haven’t seen for a year (partly because Covid). He gave me the full roster of troubles in one group of acquaintances: debt, despair, bankruptcy, divorce, death, the works.
And people are now obviously cracking up under the strain. God speed an effective vaccine.
Wait until the bills for this start to come in....everybody is going to be paying for this for the rest of their lives, and so are their kids.
What freaked me out today was a quote from a senior Swiss official (I forget where I read it, apologies) who warned his nation to expect a second wave “much much worse” than the first, including deaths.
Dunno why, but I always view the Swiss with a certain wary respect. Perhaps the most boringly sensible - or logically selfish - nation in the world.
If they really think the 2nd wave is going to be that bad - and early signs are grim - then EEEEESH
It does appear to be heading that way. Hopefully not. It could be brutal.
Does it? We’ve been entering this second wave for a month, now. Total UK critical cases, according to Worldometer data, is currently 743. Belgium - awash with new positive cases - about the same. Italy and Germany, interestingly behind us on new case numbers, have more in ICU, but not much above 1,000 each nationwide. So far, the second wave appears contagious but a lot less dangerous.
Yet we are damaging the economy with our policy response, as if it was the same.
ONS weekly deaths is doubling every 2 weeks.
Exponential is a killer. Nothing happens then everything happens at once.
Fine, but we are losing sight of the fact that the calibration is to protect the NHS from being overwhelmed, not to eliminate the virus, which - pending a vaccine - is impossible.
More people - and certainly more people who aren’t already very elderly - are dying of cancer and heart failure each week, and yet it would never enter our minds to incur this level of economic damage to drive down those deaths.
Trafalgar have just slapped a NC poll up on their site.
And you are never going to believe it, it has exactly the same demographics as the one from September.
Outstanding consistency
Likely they've got some techie who takes txt & robo push-button responses to push-polling questions (where the REAL point is message NOT the response) then re-configures results based on demographics obtained from source OTHER than the actual texts & calls.
Then Cahaly pumps it out as another Trafalgar Group "poll".
This is all supposition - but it DOES fit the known facts (including TG's own vague "explanation" re: methodology) AND is certainly right in the Lee Atwater - Karl Rove (remember him?) wheelhouse.
Nah, the polls appear g on their website are just made up. The 3 that appeared today and caused a stir are genuine-but-crap. Notice
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
Not really, it's been very disorganised and people have been proved to be irresponsible and not isolating after testing positive and the state has been left wanting in many areas. Any thoughts of exceptionalism should have been washed away by the virus response both from the state and the people.
The NHS, academic and pharmaceutical contribution is world class.
Lots of questions over the academic response over this, especially early on. Many so called experts not leveraging standard modern tools. Much that everybody laughed at the use of Excel for of the testing data, we saw with the likes of Ferguson model is was a similar shit show of coding / obselete programming language and paradigms.
We are still seeing it now with the model for how effective a circuit breaker might be providing nonsensical results.
The obsolete programming language was C, used for linux and android and therefore probably the most important and most commonly deployed language. (Obscure namedrop: I was once stuck in a lift with that there Dennis Ritchie.) I've not been following modelgate but gather some Microsofties had a go at porting Ferguson's code to C++ for no readily apparent reason except that is what programmers like to do: reimplement the wheel in a new language.
When all this is over, we might reflect that Dominic Cummings and pb's own @SeanT were right about one thing. Britain does need to invest more in research.
C is most definitely not an obsolete language it is still widely used for good reasons. That is however not to say its necessarily the best to use for those models. I can't comment on that because I haven't seen the models but C is I believe still one of the top ten languages and is still the best to choose often for embedded purposes. Like everything it is the right tool for the right purpose.
I would have more concern if the models were done in something live javascript than c.
John Carmack looked at the code and said it looked fine.
That is more than good enough for me.
Me too. The point I was making was merely that C is an old language but that doesn't make it obsolete. I wasn't arguing it was the right choice for this however I said I didn't know as hadn't seen the model.
Half the computer world still runs on C, apache web servers, linux, most embedded stuff etc.
Half of the computing world isn't trying to predict a viral replication model within a modelled real world scenario. It is what ML is made to do, using Python which has a whole host of ML libraries and can use GPU acceleration would be the correct way of doing it. I couldn't imagine anyone in my industry creating a predictive model in C and banking is at the forefront of ML, C is just old fashioned and unsuitable for using Tensorflow or PyTorch either of which are extensively used for deep learning which is what a problem like this needs.
Sigh does no one read what you write here anymore. I didn't argue c was the correct tool I merely asserted it wasn't an obsolete language. I said the right tool for the right job. I suspect in the case of covid however the model itself is the flaw rather than the language it is written in.
Further as to machine learning personally that is over hyped and unproven. We have a huge amount of examples of machine learning where what the machine learnt isn't what people thought they had taught it. Facial recognition, predictive crime etc being the two most egregious but lots of others out there.
I think the point being made is that him using C shows how outdated his approach is.
ML is great for these kinds of problems because people are predictably unpredictable. Predicting a person's behaviour is impossible, predicting population behaviour isn't and it's one of the major uses of it.
Agree that facial recognition and crime models are a pile of dogshit, but that doesn't invalidate the approach to other more suitable problems. Viral replication is probably one of the most suitable problems for ML as it's a multivariate analysis and we have a huge amount of data available now. Another good area is medical outcome modelling, a friend of mine works for Babylon and they've been doing some incredible work on it.
Don't mistake me I am not saying machine learning doesnt have a place. Sadly like much in the it world it almost feels like its becoming the silver bullet.....whatever the problem machine learning is the answer or block chain
In this case, it really is though.....something like a "Gaussian Process" is designed exactly for these type of problems i.e. something with a small amount of noisy data, where you have some idea about the range of the model parameters and you want to be able to make predictions into the future which by construction provide measurements of uncertainty (something a deep neural net can't do).
Trafalgar have just slapped a NC poll up on their site.
And you are never going to believe it, it has exactly the same demographics as the one from September.
Outstanding consistency
Likely they've got some techie who takes txt & robo push-button responses to push-polling questions (where the REAL point is message NOT the response) then re-configures results based on demographics obtained from source OTHER than the actual texts & calls.
Then Cahaly pumps it out as another Trafalgar Group "poll".
This is all supposition - but it DOES fit the known facts (including TG's own vague "explanation" re: methodology) AND is certainly right in the Lee Atwater - Karl Rove (remember him?) wheelhouse.
Nah, the polls appear g on their website are just made up. The 3 that appeared today and caused a stir are genuine-but-crap. Notice
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
Not really, it's been very disorganised and people have been proved to be irresponsible and not isolating after testing positive and the state has been left wanting in many areas. Any thoughts of exceptionalism should have been washed away by the virus response both from the state and the people.
The NHS, academic and pharmaceutical contribution is world class.
Lots of questions over the academic response over this, especially early on. Many so called experts not leveraging standard modern tools. Much that everybody laughed at the use of Excel for of the testing data, we saw with the likes of Ferguson model is was a similar shit show of coding / obselete programming language and paradigms.
We are still seeing it now with the model for how effective a circuit breaker might be providing nonsensical results.
The obsolete programming language was C, used for linux and android and therefore probably the most important and most commonly deployed language. (Obscure namedrop: I was once stuck in a lift with that there Dennis Ritchie.) I've not been following modelgate but gather some Microsofties had a go at porting Ferguson's code to C++ for no readily apparent reason except that is what programmers like to do: reimplement the wheel in a new language.
When all this is over, we might reflect that Dominic Cummings and pb's own @SeanT were right about one thing. Britain does need to invest more in research.
C is most definitely not an obsolete language it is still widely used for good reasons. That is however not to say its necessarily the best to use for those models. I can't comment on that because I haven't seen the models but C is I believe still one of the top ten languages and is still the best to choose often for embedded purposes. Like everything it is the right tool for the right purpose.
I would have more concern if the models were done in something live javascript than c.
John Carmack looked at the code and said it looked fine.
That is more than good enough for me.
Me too. The point I was making was merely that C is an old language but that doesn't make it obsolete. I wasn't arguing it was the right choice for this however I said I didn't know as hadn't seen the model.
Half the computer world still runs on C, apache web servers, linux, most embedded stuff etc.
Half of the computing world isn't trying to predict a viral replication model within a modelled real world scenario. It is what ML is made to do, using Python which has a whole host of ML libraries and can use GPU acceleration would be the correct way of doing it. I couldn't imagine anyone in my industry creating a predictive model in C and banking is at the forefront of ML, C is just old fashioned and unsuitable for using Tensorflow or PyTorch either of which are extensively used for deep learning which is what a problem like this needs.
Sigh does no one read what you write here anymore. I didn't argue c was the correct tool I merely asserted it wasn't an obsolete language. I said the right tool for the right job. I suspect in the case of covid however the model itself is the flaw rather than the language it is written in.
Further as to machine learning personally that is over hyped and unproven. We have a huge amount of examples of machine learning where what the machine learnt isn't what people thought they had taught it. Facial recognition, predictive crime etc being the two most egregious but lots of others out there.
I think the point being made is that him using C shows how outdated his approach is.
ML is great for these kinds of problems because people are predictably unpredictable. Predicting a person's behaviour is impossible, predicting population behaviour isn't and it's one of the major uses of it.
Agree that facial recognition and crime models are a pile of dogshit, but that doesn't invalidate the approach to other more suitable problems. Viral replication is probably one of the most suitable problems for ML as it's a multivariate analysis and we have a huge amount of data available now. Another good area is medical outcome modelling, a friend of mine works for Babylon and they've been doing some incredible work on it.
Don't mistake me I am not saying machine learning doesnt have a place. Sadly like much in the it world it almost feels like its becoming the silver bullet.....whatever the problem machine learning is the answer or block chain
I'm not so sure, I've implemented ML in a few areas of my work and it's been very helpful, but I agree that it isn't a silver bullet for every scenario. It's definitely well ahead of Blockchain in real world use, I mean in gaming DLSS is about to make native resolution obsolete and that's built on ML. Blockchain is vaporware, ML definitely isn't.
One by one, Trump’s friends and allies peel away as they face up to the increasingly inevitable. Beyond US shores, soon only Bozo and HY will be cheering him on...
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
Not really, it's been very disorganised and people have been proved to be irresponsible and not isolating after testing positive and the state has been left wanting in many areas. Any thoughts of exceptionalism should have been washed away by the virus response both from the state and the people.
The NHS, academic and pharmaceutical contribution is world class.
Lots of questions over the academic response over this, especially early on. Many so called experts not leveraging standard modern tools. Much that everybody laughed at the use of Excel for of the testing data, we saw with the likes of Ferguson model is was a similar shit show of coding / obselete programming language and paradigms.
We are still seeing it now with the model for how effective a circuit breaker might be providing nonsensical results.
The obsolete programming language was C, used for linux and android and therefore probably the most important and most commonly deployed language. (Obscure namedrop: I was once stuck in a lift with that there Dennis Ritchie.) I've not been following modelgate but gather some Microsofties had a go at porting Ferguson's code to C++ for no readily apparent reason except that is what programmers like to do: reimplement the wheel in a new language.
When all this is over, we might reflect that Dominic Cummings and pb's own @SeanT were right about one thing. Britain does need to invest more in research.
Firstly, a portion of the model was auto-generated code from Fortran. Fortran...what did he code it on, a 486?
Secondly, nobody who knows what they are doing does statistical modelling in C these days...
Furthermore, it was single threaded, made no use of the likes of Intel MKL, Cuda or basically any modern programming suite of tools / libraries that people use for modelling complex problems.
Given how simple the model is fundamentally, going parallel would have been pointless.
My point is today anybody who knows what they are doing doesn't have to make that as a big design decision. Even if you aren't a coding whizz, it is trivial to make multi-threaded, use GPU, etc from the get go and the use of something like MKL (or a library built on it) is standard if you are considering doing mathematical modelling.
From talking to some people who were peripherally involved in reimplementing the model (multiple different groups, multiple languages) - it runs fine single threaded.
The interesting bit would be getting the ML "thinkers" to have a look at the whole area - looks like that wasn't in Ferguson's wheelhouse.
I think we have got rather bogged down in that it runs fine single threaded. The point is nobody with modern knowledge of state of the art mathematical modelling would approach it like this. And the ability to use multi-threading and distributed GPU processing for starters allows you to create more complex models, not simply its runs in hours rather than minutes.
Welcome back @LadyG - hope you managed your road trip safely.
Ta. I’m still on it. So far it’s been a mixture of highly enjoyable and very sobering. Saw lots of family, had quite jolly times, made the most of Cornish sun and good seafood. But today I met a friend in Devon who I haven’t seen for a year (partly because Covid). He gave me the full roster of troubles in one group of acquaintances: debt, despair, bankruptcy, divorce, death, the works.
And people are now obviously cracking up under the strain. God speed an effective vaccine.
Wait until the bills for this start to come in....everybody is going to be paying for this for the rest of their lives, and so are their kids.
What freaked me out today was a quote from a senior Swiss official (I forget where I read it, apologies) who warned his nation to expect a second wave “much much worse” than the first, including deaths.
Dunno why, but I always view the Swiss with a certain wary respect. Perhaps the most boringly sensible - or logically selfish - nation in the world.
If they really think the 2nd wave is going to be that bad - and early signs are grim - then EEEEESH
It does appear to be heading that way. Hopefully not. It could be brutal.
Does it? We’ve been entering this second wave for a month, now. Total UK critical cases, according to Worldometer data, is currently 743. Belgium - awash with new positive cases - about the same. Italy and Germany, interestingly behind us on new case numbers, have more in ICU, but not much above 1,000 each nationwide. So far, the second wave appears contagious but a lot less dangerous.
Yet we are damaging the economy with our policy response, as if it was the same.
ONS weekly deaths is doubling every 2 weeks.
Exponential is a killer. Nothing happens then everything happens at once.
Fine, but we are losing sight of the fact that the calibration is to protect the NHS from being overwhelmed, not to eliminate the virus, which - pending a vaccine - is impossible.
More people - and certainly more people who aren’t already very elderly - are dying of cancer and heart failure each week, and yet it would never enter our minds to incur this level of economic damage to drive down those deaths.
Yes but neither will those conditions double in mortality over less than a fortnight.
For quite a while one of the best COVID-19 models was one guy from MIT applying machine learning to an off-the-shelf model. He was consitently beating all sorts of fancy national labs and universities, many of which were so crap at modelling that they could not even beat the baseline of repeating the previous measured change. To be fair modelling has improved dramatically over the last six months or so, but I don't think it's unreasonable to state that the bulk of epidemiological modelling when the pandemic hit was a load of rubbish.
Can we have a link please to the work of this guy from MIT ? Thanks.
Sadly he's stopped running his model (given the amount of work he's done I should think he needs a rest), but there is a hell of a lot of good stuff to read on his site. It's well worth looking at his ranking for the various US models, a lot of very prominent organisations have been pumping out absolute rubbish for months on end.
Check out the first big colour-coded table for comparisons on the page below.
Incidentally I am in a very pleasant hotel in Bristol - indeed, The Bristol Hotel.
It is highly covid cautious in its check-in procedures, elevator protocols, room cleaning, etc.
Yet the restaurant is absolutely packed and rocking with maskless people quaffing wine and expelling aerosols all over the shop. There are lots of bars and pizzerias and sushi bars all around, and they are similarly busy. It’s very upbeat.... but I just don’t see how we avoid following Italy, France, Spain, etc, into a pretty severe second lockdown and early curfews. Britain is not exceptionally immune. Just late to the party. Again.
On the upside Bristol - a city I barely knew - turns out to be highly beguiling. A hidden gem. When the Wuhan Flu is over I might return to buy a plague-vacated Georgian house for £10.
That's a brilliantly designed tweet by Marcus Rashford there, which has been written to appeal to both the left and the right. And it does so very well indeed.
Yes, he might have some people advising him on this but I can easily believe he's savvy enough to work it out for himself and, even if not, shrewd enough to listen to those who are.
The second part is they key. It doesn't matter all that much how much is him and how much is others, listening to good advice is a skill.
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
Not really, it's been very disorganised and people have been proved to be irresponsible and not isolating after testing positive and the state has been left wanting in many areas. Any thoughts of exceptionalism should have been washed away by the virus response both from the state and the people.
The NHS, academic and pharmaceutical contribution is world class.
Lots of questions over the academic response over this, especially early on. Many so called experts not leveraging standard modern tools. Much that everybody laughed at the use of Excel for of the testing data, we saw with the likes of Ferguson model is was a similar shit show of coding / obselete programming language and paradigms.
We are still seeing it now with the model for how effective a circuit breaker might be providing nonsensical results.
The obsolete programming language was C, used for linux and android and therefore probably the most important and most commonly deployed language. (Obscure namedrop: I was once stuck in a lift with that there Dennis Ritchie.) I've not been following modelgate but gather some Microsofties had a go at porting Ferguson's code to C++ for no readily apparent reason except that is what programmers like to do: reimplement the wheel in a new language.
When all this is over, we might reflect that Dominic Cummings and pb's own @SeanT were right about one thing. Britain does need to invest more in research.
Firstly, a portion of the model was auto-generated code from Fortran. Fortran...what did he code it on, a 486?
Secondly, nobody who knows what they are doing does statistical modelling in C these days...
Furthermore, it was single threaded, made no use of the likes of Intel MKL, Cuda or basically any modern programming suite of tools / libraries that people use for modelling complex problems.
Given how simple the model is fundamentally, going parallel would have been pointless.
My point is today anybody who knows what they are doing doesn't have to make that as a big design decision. Even if you aren't a coding whizz, it is trivial to make multi-threaded, use GPU, etc from the get go and the use of something like MKL (or a library built on it) is standard if you are considering doing mathematical modelling.
From talking to some people who were peripherally involved in reimplementing the model (multiple different groups, multiple languages) - it runs fine single threaded.
The interesting bit would be getting the ML "thinkers" to have a look at the whole area - looks like that wasn't in Ferguson's wheelhouse.
Getting ML people involved wouldn't be difficult, just recruit them, 3-5 data engineers, 10-12 data analysts and then 2 or 3 data scientists. Stick everything on GCP. It's basically what every bank has done in the last few years. With data teams expanding from there once they've proved their value.
Trafalgar have just slapped a NC poll up on their site.
And you are never going to believe it, it has exactly the same demographics as the one from September.
Outstanding consistency
Likely they've got some techie who takes txt & robo push-button responses to push-polling questions (where the REAL point is message NOT the response) then re-configures results based on demographics obtained from source OTHER than the actual texts & calls.
Then Cahaly pumps it out as another Trafalgar Group "poll".
This is all supposition - but it DOES fit the known facts (including TG's own vague "explanation" re: methodology) AND is certainly right in the Lee Atwater - Karl Rove (remember him?) wheelhouse.
Nah, the polls appear g on their website are just made up. The 3 that appeared today and caused a stir are genuine-but-crap. Notice
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
Not really, it's been very disorganised and people have been proved to be irresponsible and not isolating after testing positive and the state has been left wanting in many areas. Any thoughts of exceptionalism should have been washed away by the virus response both from the state and the people.
The NHS, academic and pharmaceutical contribution is world class.
Lots of questions over the academic response over this, especially early on. Many so called experts not leveraging standard modern tools. Much that everybody laughed at the use of Excel for of the testing data, we saw with the likes of Ferguson model is was a similar shit show of coding / obselete programming language and paradigms.
We are still seeing it now with the model for how effective a circuit breaker might be providing nonsensical results.
The obsolete programming language was C, used for linux and android and therefore probably the most important and most commonly deployed language. (Obscure namedrop: I was once stuck in a lift with that there Dennis Ritchie.) I've not been following modelgate but gather some Microsofties had a go at porting Ferguson's code to C++ for no readily apparent reason except that is what programmers like to do: reimplement the wheel in a new language.
When all this is over, we might reflect that Dominic Cummings and pb's own @SeanT were right about one thing. Britain does need to invest more in research.
C is most definitely not an obsolete language it is still widely used for good reasons. That is however not to say its necessarily the best to use for those models. I can't comment on that because I haven't seen the models but C is I believe still one of the top ten languages and is still the best to choose often for embedded purposes. Like everything it is the right tool for the right purpose.
I would have more concern if the models were done in something live javascript than c.
John Carmack looked at the code and said it looked fine.
That is more than good enough for me.
Me too. The point I was making was merely that C is an old language but that doesn't make it obsolete. I wasn't arguing it was the right choice for this however I said I didn't know as hadn't seen the model.
Half the computer world still runs on C, apache web servers, linux, most embedded stuff etc.
Half of the computing world isn't trying to predict a viral replication model within a modelled real world scenario. It is what ML is made to do, using Python which has a whole host of ML libraries and can use GPU acceleration would be the correct way of doing it. I couldn't imagine anyone in my industry creating a predictive model in C and banking is at the forefront of ML, C is just old fashioned and unsuitable for using Tensorflow or PyTorch either of which are extensively used for deep learning which is what a problem like this needs.
Sigh does no one read what you write here anymore. I didn't argue c was the correct tool I merely asserted it wasn't an obsolete language. I said the right tool for the right job. I suspect in the case of covid however the model itself is the flaw rather than the language it is written in.
Further as to machine learning personally that is over hyped and unproven. We have a huge amount of examples of machine learning where what the machine learnt isn't what people thought they had taught it. Facial recognition, predictive crime etc being the two most egregious but lots of others out there.
I think the point being made is that him using C shows how outdated his approach is.
ML is great for these kinds of problems because people are predictably unpredictable. Predicting a person's behaviour is impossible, predicting population behaviour isn't and it's one of the major uses of it.
Agree that facial recognition and crime models are a pile of dogshit, but that doesn't invalidate the approach to other more suitable problems. Viral replication is probably one of the most suitable problems for ML as it's a multivariate analysis and we have a huge amount of data available now. Another good area is medical outcome modelling, a friend of mine works for Babylon and they've been doing some incredible work on it.
Don't mistake me I am not saying machine learning doesnt have a place. Sadly like much in the it world it almost feels like its becoming the silver bullet.....whatever the problem machine learning is the answer or block chain
I'm not so sure, I've implemented ML in a few areas of my work and it's been very helpful, but I agree that it isn't a silver bullet for every scenario. It's definitely well ahead of Blockchain in real world use, I mean in gaming DLSS is about to make native resolution obsolete and that's built on ML. Blockchain is vaporware, ML definitely isn't.
I think the issue with ML is a bit like some of the government announcements, overpromising. People seem to want to claim it is the magic bullet and will solve problems to 99.9% accuracy. When it is better to say, no, but using modern ML approaches can produce results which are superior to older techniques.
e.g. For computer vision, 10-15 years ago, everybody tried to manually come up with some set of descriptors to find particular objects (or parts of objects in a scene). That approach never produced robust or generalizable solutions.
Now we have models that have learned to identify objects with a much improved level of accuracy. They can't do it 100%, they can be tricked, but in comparison to state of the art from 15 years ago, it is chalk and cheese, where we are at a stage that they models can be built into real world applications (with the knowledge they won't work 100% of the time).
Now if you want to debate if some variant of a CNN, ResNet or GAN is THE solution, now that is definitely where there is huge over-promising. They aren't, but they have shown that we can harness the incredible power of modern CPU / GPUs to process enormous amounts of data to learn a model of a problem using latent variants, which humans are incapable of doing.
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
Not really, it's been very disorganised and people have been proved to be irresponsible and not isolating after testing positive and the state has been left wanting in many areas. Any thoughts of exceptionalism should have been washed away by the virus response both from the state and the people.
The NHS, academic and pharmaceutical contribution is world class.
Lots of questions over the academic response over this, especially early on. Many so called experts not leveraging standard modern tools. Much that everybody laughed at the use of Excel for of the testing data, we saw with the likes of Ferguson model is was a similar shit show of coding / obselete programming language and paradigms.
We are still seeing it now with the model for how effective a circuit breaker might be providing nonsensical results.
The obsolete programming language was C, used for linux and android and therefore probably the most important and most commonly deployed language. (Obscure namedrop: I was once stuck in a lift with that there Dennis Ritchie.) I've not been following modelgate but gather some Microsofties had a go at porting Ferguson's code to C++ for no readily apparent reason except that is what programmers like to do: reimplement the wheel in a new language.
When all this is over, we might reflect that Dominic Cummings and pb's own @SeanT were right about one thing. Britain does need to invest more in research.
Firstly, a portion of the model was auto-generated code from Fortran. Fortran...what did he code it on, a 486?
Secondly, nobody who knows what they are doing does statistical modelling in C these days...
Furthermore, it was single threaded, made no use of the likes of Intel MKL, Cuda or basically any modern programming suite of tools / libraries that people use for modelling complex problems.
Given how simple the model is fundamentally, going parallel would have been pointless.
My point is today anybody who knows what they are doing doesn't have to make that as a big design decision. Even if you aren't a coding whizz, it is trivial to make multi-threaded, use GPU, etc from the get go and the use of something like MKL (or a library built on it) is standard if you are considering doing mathematical modelling.
From talking to some people who were peripherally involved in reimplementing the model (multiple different groups, multiple languages) - it runs fine single threaded.
The interesting bit would be getting the ML "thinkers" to have a look at the whole area - looks like that wasn't in Ferguson's wheelhouse.
I think we have got rather bogged down in that it runs fine single threaded. The point is nobody with modern knowledge of state of the art mathematical modelling would approach it like this. And the ability to use multi-threading and distributed GPU processing for starters allows you to create more complex models, not simply its runs in hours rather than minutes.
Yes - the problem is not the technology. It's the thinking. Reminds me of trying to get quants to write their algorithms as matrix maths.....
One by one, Trump’s friends and allies peel away as they face up to the increasingly inevitable. Beyond US shores, soon only Bozo and HY will be cheering him on...
Putin has NOT stopped cheering for his fellow Putinist. But he DOES understand that his OVERT support would be toxic for Trumpsky.
Beyond that, is trying to establish some basis of relationship with (increasingly likely) Biden administration.
I'm not so sure, I've implemented ML in a few areas of my work and it's been very helpful, but I agree that it isn't a silver bullet for every scenario. It's definitely well ahead of Blockchain in real world use, I mean in gaming DLSS is about to make native resolution obsolete and that's built on ML. Blockchain is vaporware, ML definitely isn't.
As with a lot of things the best solution is the tool you know and have to hand which was why the original prediction was written in C (and that still just a gout made sense when it was written).
One by one, Trump’s friends and allies peel away as they face up to the increasingly inevitable. Beyond US shores, soon only Bozo and HY will be cheering him on...
I think you have the wrong end of the stick, the Russians are heavily involved in Burisma (the Cypriot connection should have given that away), they really don't want that to blow up.
Which btw, should tell you everything you need to know about the hypocrisy of the Democrats talking about Trump and Russian interference.
Tesco Pengam Green will be like Harrod's January sale as the doors open at 9 am on New Year's day.
In that article Drakeford says
'I won't need - I don't think - to buy clothing over this two weeks and I think many many people in Wales will be in that poition to. For me it won't be essential but I recognise some people for entirely unexpected reasons which they could not have foreseen will need to buy items'
So laws in Wales are being made by a first minister who does not think
Trafalgar have just slapped a NC poll up on their site.
And you are never going to believe it, it has exactly the same demographics as the one from September.
Outstanding consistency
Likely they've got some techie who takes txt & robo push-button responses to push-polling questions (where the REAL point is message NOT the response) then re-configures results based on demographics obtained from source OTHER than the actual texts & calls.
Then Cahaly pumps it out as another Trafalgar Group "poll".
This is all supposition - but it DOES fit the known facts (including TG's own vague "explanation" re: methodology) AND is certainly right in the Lee Atwater - Karl Rove (remember him?) wheelhouse.
Nah, the polls appear g on their website are just made up. The 3 that appeared today and caused a stir are genuine-but-crap. Notice
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
Not really, it's been very disorganised and people have been proved to be irresponsible and not isolating after testing positive and the state has been left wanting in many areas. Any thoughts of exceptionalism should have been washed away by the virus response both from the state and the people.
The NHS, academic and pharmaceutical contribution is world class.
Lots of questions over the academic response over this, especially early on. Many so called experts not leveraging standard modern tools. Much that everybody laughed at the use of Excel for of the testing data, we saw with the likes of Ferguson model is was a similar shit show of coding / obselete programming language and paradigms.
We are still seeing it now with the model for how effective a circuit breaker might be providing nonsensical results.
The obsolete programming language was C, used for linux and android and therefore probably the most important and most commonly deployed language. (Obscure namedrop: I was once stuck in a lift with that there Dennis Ritchie.) I've not been following modelgate but gather some Microsofties had a go at porting Ferguson's code to C++ for no readily apparent reason except that is what programmers like to do: reimplement the wheel in a new language.
When all this is over, we might reflect that Dominic Cummings and pb's own @SeanT were right about one thing. Britain does need to invest more in research.
C is most definitely not an obsolete language it is still widely used for good reasons. That is however not to say its necessarily the best to use for those models. I can't comment on that because I haven't seen the models but C is I believe still one of the top ten languages and is still the best to choose often for embedded purposes. Like everything it is the right tool for the right purpose.
I would have more concern if the models were done in something live javascript than c.
John Carmack looked at the code and said it looked fine.
That is more than good enough for me.
Me too. The point I was making was merely that C is an old language but that doesn't make it obsolete. I wasn't arguing it was the right choice for this however I said I didn't know as hadn't seen the model.
Half the computer world still runs on C, apache web servers, linux, most embedded stuff etc.
Half of the computing world isn't trying to predict a viral replication model within a modelled real world scenario. It is what ML is made to do, using Python which has a whole host of ML libraries and can use GPU acceleration would be the correct way of doing it. I couldn't imagine anyone in my industry creating a predictive model in C and banking is at the forefront of ML, C is just old fashioned and unsuitable for using Tensorflow or PyTorch either of which are extensively used for deep learning which is what a problem like this needs.
Sigh does no one read what you write here anymore. I didn't argue c was the correct tool I merely asserted it wasn't an obsolete language. I said the right tool for the right job. I suspect in the case of covid however the model itself is the flaw rather than the language it is written in.
Further as to machine learning personally that is over hyped and unproven. We have a huge amount of examples of machine learning where what the machine learnt isn't what people thought they had taught it. Facial recognition, predictive crime etc being the two most egregious but lots of others out there.
I think the point being made is that him using C shows how outdated his approach is.
ML is great for these kinds of problems because people are predictably unpredictable. Predicting a person's behaviour is impossible, predicting population behaviour isn't and it's one of the major uses of it.
Agree that facial recognition and crime models are a pile of dogshit, but that doesn't invalidate the approach to other more suitable problems. Viral replication is probably one of the most suitable problems for ML as it's a multivariate analysis and we have a huge amount of data available now. Another good area is medical outcome modelling, a friend of mine works for Babylon and they've been doing some incredible work on it.
Don't mistake me I am not saying machine learning doesnt have a place. Sadly like much in the it world it almost feels like its becoming the silver bullet.....whatever the problem machine learning is the answer or block chain
I'm not so sure, I've implemented ML in a few areas of my work and it's been very helpful, but I agree that it isn't a silver bullet for every scenario. It's definitely well ahead of Blockchain in real world use, I mean in gaming DLSS is about to make native resolution obsolete and that's built on ML. Blockchain is vaporware, ML definitely isn't.
By machine learning, are you talking about neural networks, or genetic algorithms? The former require tons of historical data, so are not so useful if you're dealing with inherently unknowable systems or situations where you either have no data or you only have low confidence in what you have. I think, for example, they will axiomatically fail to predict black swans or emergent properties that are not in the historical data used to train them.
For genetic algorithms, I am not sure how evolved solutions in silico will necessarily be better than those evolved through acquisition of genuine expertise.
I am not an expert in this area, and so would welcome correction and enlightenment if I am wrong.
PS These comments relate to machine learning in complex adaptive systems - I realize that it can be very effective in better defined, deterministic systems
I presume Corbyn will be organizing a protest to complain about the disproportionate and excessive force used by our special forces and how we should have sat down for a cup of tea to better understand these poor migrants.
Love this quote from a foreign election - our parties think they have difficulties in campaign season Mr Faure, who had inherited power from his predecessor four years ago, was unable to distance his party's campaign from mounting evidence of past political murders, torture and corruption when Seychelles was still a one-party state.
Trafalgar have just slapped a NC poll up on their site.
And you are never going to believe it, it has exactly the same demographics as the one from September.
Outstanding consistency
Likely they've got some techie who takes txt & robo push-button responses to push-polling questions (where the REAL point is message NOT the response) then re-configures results based on demographics obtained from source OTHER than the actual texts & calls.
Then Cahaly pumps it out as another Trafalgar Group "poll".
This is all supposition - but it DOES fit the known facts (including TG's own vague "explanation" re: methodology) AND is certainly right in the Lee Atwater - Karl Rove (remember him?) wheelhouse.
Nah, the polls appear g on their website are just made up. The 3 that appeared today and caused a stir are genuine-but-crap. Notice
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
Not really, it's been very disorganised and people have been proved to be irresponsible and not isolating after testing positive and the state has been left wanting in many areas. Any thoughts of exceptionalism should have been washed away by the virus response both from the state and the people.
The NHS, academic and pharmaceutical contribution is world class.
Lots of questions over the academic response over this, especially early on. Many so called experts not leveraging standard modern tools. Much that everybody laughed at the use of Excel for of the testing data, we saw with the likes of Ferguson model is was a similar shit show of coding / obselete programming language and paradigms.
We are still seeing it now with the model for how effective a circuit breaker might be providing nonsensical results.
The obsolete programming language was C, used for linux and android and therefore probably the most important and most commonly deployed language. (Obscure namedrop: I was once stuck in a lift with that there Dennis Ritchie.) I've not been following modelgate but gather some Microsofties had a go at porting Ferguson's code to C++ for no readily apparent reason except that is what programmers like to do: reimplement the wheel in a new language.
When all this is over, we might reflect that Dominic Cummings and pb's own @SeanT were right about one thing. Britain does need to invest more in research.
C is most definitely not an obsolete language it is still widely used for good reasons. That is however not to say its necessarily the best to use for those models. I can't comment on that because I haven't seen the models but C is I believe still one of the top ten languages and is still the best to choose often for embedded purposes. Like everything it is the right tool for the right purpose.
I would have more concern if the models were done in something live javascript than c.
John Carmack looked at the code and said it looked fine.
That is more than good enough for me.
Me too. The point I was making was merely that C is an old language but that doesn't make it obsolete. I wasn't arguing it was the right choice for this however I said I didn't know as hadn't seen the model.
Half the computer world still runs on C, apache web servers, linux, most embedded stuff etc.
Half of the computing world isn't trying to predict a viral replication model within a modelled real world scenario. It is what ML is made to do, using Python which has a whole host of ML libraries and can use GPU acceleration would be the correct way of doing it. I couldn't imagine anyone in my industry creating a predictive model in C and banking is at the forefront of ML, C is just old fashioned and unsuitable for using Tensorflow or PyTorch either of which are extensively used for deep learning which is what a problem like this needs.
Sigh does no one read what you write here anymore. I didn't argue c was the correct tool I merely asserted it wasn't an obsolete language. I said the right tool for the right job. I suspect in the case of covid however the model itself is the flaw rather than the language it is written in.
Further as to machine learning personally that is over hyped and unproven. We have a huge amount of examples of machine learning where what the machine learnt isn't what people thought they had taught it. Facial recognition, predictive crime etc being the two most egregious but lots of others out there.
I think the point being made is that him using C shows how outdated his approach is.
ML is great for these kinds of problems because people are predictably unpredictable. Predicting a person's behaviour is impossible, predicting population behaviour isn't and it's one of the major uses of it.
Agree that facial recognition and crime models are a pile of dogshit, but that doesn't invalidate the approach to other more suitable problems. Viral replication is probably one of the most suitable problems for ML as it's a multivariate analysis and we have a huge amount of data available now. Another good area is medical outcome modelling, a friend of mine works for Babylon and they've been doing some incredible work on it.
Don't mistake me I am not saying machine learning doesnt have a place. Sadly like much in the it world it almost feels like its becoming the silver bullet.....whatever the problem machine learning is the answer or block chain
I'm not so sure, I've implemented ML in a few areas of my work and it's been very helpful, but I agree that it isn't a silver bullet for every scenario. It's definitely well ahead of Blockchain in real world use, I mean in gaming DLSS is about to make native resolution obsolete and that's built on ML. Blockchain is vaporware, ML definitely isn't.
By machine learning, are you talking about neural networks, or genetic algorithms? The former require tons of historical data, so are not so useful if you're dealing with inherently unknowable systems or situations where you either have no data or you only have low confidence in what you have. I think, for example, they will axiomatically fail to predict black swans or emergent properties that are not in the historical data used to train them.
For genetic algorithms, I am not sure how evolved solutions in silico will necessarily be better than those evolved through acquisition of genuine expertise.
I am not an expert in this area, and so would welcome correction and enlightenment if I am wrong.
PS These comments relate to machine learning in complex adaptive systems - I realize that it can be very effective in better defined, deterministic systems
Neither a Deep Neural Net or a genetic algorithm is really the best approach. Something like Gaussian Processes based model is (and no that doesn't mean just fit a gaussian curve though some data).
I'm not so sure, I've implemented ML in a few areas of my work and it's been very helpful, but I agree that it isn't a silver bullet for every scenario. It's definitely well ahead of Blockchain in real world use, I mean in gaming DLSS is about to make native resolution obsolete and that's built on ML. Blockchain is vaporware, ML definitely isn't.
As with a lot of things the best solution is the tool you know and have to hand which was why the original prediction was written in C (and that still just a gout made sense when it was written).
And that's the criticism, the best model we have isn't the best available. What our academics know isn't the best available predictive analytics. You simply can't built a complex ML model in C in the same way you could in Python and you wouldn't be able to train it properly without multi threading, it would take far too long.
One by one, Trump’s friends and allies peel away as they face up to the increasingly inevitable. Beyond US shores, soon only Bozo and HY will be cheering him on...
I think you have the wrong end of the stick, the Russians are heavily involved in Burisma (the Cypriot connection should have given that away), they really don't want that to blow up.
Which btw, should tell you everything you need to know about the hypocrisy of the Democrats talking about Trump and Russian interference.
I'm sorry, I don't speak fringe-right-wing-conspiracy. Can you translate into English?
Trafalgar have just slapped a NC poll up on their site.
And you are never going to believe it, it has exactly the same demographics as the one from September.
Outstanding consistency
Likely they've got some techie who takes txt & robo push-button responses to push-polling questions (where the REAL point is message NOT the response) then re-configures results based on demographics obtained from source OTHER than the actual texts & calls.
Then Cahaly pumps it out as another Trafalgar Group "poll".
This is all supposition - but it DOES fit the known facts (including TG's own vague "explanation" re: methodology) AND is certainly right in the Lee Atwater - Karl Rove (remember him?) wheelhouse.
Nah, the polls appear g on their website are just made up. The 3 that appeared today and caused a stir are genuine-but-crap. Notice
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
Not really, it's been very disorganised and people have been proved to be irresponsible and not isolating after testing positive and the state has been left wanting in many areas. Any thoughts of exceptionalism should have been washed away by the virus response both from the state and the people.
The NHS, academic and pharmaceutical contribution is world class.
Lots of questions over the academic response over this, especially early on. Many so called experts not leveraging standard modern tools. Much that everybody laughed at the use of Excel for of the testing data, we saw with the likes of Ferguson model is was a similar shit show of coding / obselete programming language and paradigms.
We are still seeing it now with the model for how effective a circuit breaker might be providing nonsensical results.
The obsolete programming language was C, used for linux and android and therefore probably the most important and most commonly deployed language. (Obscure namedrop: I was once stuck in a lift with that there Dennis Ritchie.) I've not been following modelgate but gather some Microsofties had a go at porting Ferguson's code to C++ for no readily apparent reason except that is what programmers like to do: reimplement the wheel in a new language.
When all this is over, we might reflect that Dominic Cummings and pb's own @SeanT were right about one thing. Britain does need to invest more in research.
C is most definitely not an obsolete language it is still widely used for good reasons. That is however not to say its necessarily the best to use for those models. I can't comment on that because I haven't seen the models but C is I believe still one of the top ten languages and is still the best to choose often for embedded purposes. Like everything it is the right tool for the right purpose.
I would have more concern if the models were done in something live javascript than c.
John Carmack looked at the code and said it looked fine.
That is more than good enough for me.
Me too. The point I was making was merely that C is an old language but that doesn't make it obsolete. I wasn't arguing it was the right choice for this however I said I didn't know as hadn't seen the model.
Half the computer world still runs on C, apache web servers, linux, most embedded stuff etc.
Half of the computing world isn't trying to predict a viral replication model within a modelled real world scenario. It is what ML is made to do, using Python which has a whole host of ML libraries and can use GPU acceleration would be the correct way of doing it. I couldn't imagine anyone in my industry creating a predictive model in C and banking is at the forefront of ML, C is just old fashioned and unsuitable for using Tensorflow or PyTorch either of which are extensively used for deep learning which is what a problem like this needs.
Sigh does no one read what you write here anymore. I didn't argue c was the correct tool I merely asserted it wasn't an obsolete language. I said the right tool for the right job. I suspect in the case of covid however the model itself is the flaw rather than the language it is written in.
Further as to machine learning personally that is over hyped and unproven. We have a huge amount of examples of machine learning where what the machine learnt isn't what people thought they had taught it. Facial recognition, predictive crime etc being the two most egregious but lots of others out there.
I think the point being made is that him using C shows how outdated his approach is.
ML is great for these kinds of problems because people are predictably unpredictable. Predicting a person's behaviour is impossible, predicting population behaviour isn't and it's one of the major uses of it.
Agree that facial recognition and crime models are a pile of dogshit, but that doesn't invalidate the approach to other more suitable problems. Viral replication is probably one of the most suitable problems for ML as it's a multivariate analysis and we have a huge amount of data available now. Another good area is medical outcome modelling, a friend of mine works for Babylon and they've been doing some incredible work on it.
Don't mistake me I am not saying machine learning doesnt have a place. Sadly like much in the it world it almost feels like its becoming the silver bullet.....whatever the problem machine learning is the answer or block chain
I'm not so sure, I've implemented ML in a few areas of my work and it's been very helpful, but I agree that it isn't a silver bullet for every scenario. It's definitely well ahead of Blockchain in real world use, I mean in gaming DLSS is about to make native resolution obsolete and that's built on ML. Blockchain is vaporware, ML definitely isn't.
I think the issue with ML is a bit like some of the government announcements, overpromising. People seem to want to claim it is the magic bullet and will solve problems to 99.9% accuracy. When it is better to say, no, but using modern ML approaches can produce results which are superior to older techniques.
e.g. For computer vision, 10-15 years ago, everybody tried to manually come up with some set of descriptors to find particular objects (or parts of objects in a scene). That approach never produced robust or generalizable solutions.
Now we have models that have learned to identify objects with a much improved level of accuracy. They can't do it 100%, they can be tricked, but in comparison to state of the art from 15 years ago, it is chalk and cheese, where we are at a stage that they models can be built into real world applications (with the knowledge they won't work 100% of the time).
Now if you want to debate if some variant of a CNN, ResNet or GAN is THE solution, now that is definitely where there is huge over-promising. They aren't, but they have shown that we can harness the incredible power of modern CPU / GPUs to process enormous amounts of data to learn a model of a problem using latent variants, which humans are incapable of doing.
As an aside the interesting side of ml that rarely gets mentioned widely is people finding ways to fool machine learning especially around computer vision. I have noted stickers and tshirts and other clothing already being sold which are claimed to confuse it.
It is seriously worrying right across Europe just now
I'm not really sure why the big Western European nations (excluding Germany) have all been about as bad as one another.
Because they are culturally, geographically, economically and demographically very similar?
But most of Europe have 2-3 of those and aren't even close.
Most of Europe is now equalising on its covid outcomes. IE nations that did “well” in the first wave - Czechia, Poland - are now doing badly in the second. In the end it may all be a wash, bar a few outliers. Germany is not immune, either.
Not really. In the US, the parties only publish polls that show themselves well ahead. In the UK, party published (or created) polls always show that it’s too close to call!
I'm not so sure, I've implemented ML in a few areas of my work and it's been very helpful, but I agree that it isn't a silver bullet for every scenario. It's definitely well ahead of Blockchain in real world use, I mean in gaming DLSS is about to make native resolution obsolete and that's built on ML. Blockchain is vaporware, ML definitely isn't.
As with a lot of things the best solution is the tool you know and have to hand which was why the original prediction was written in C (and that still just a gout made sense when it was written).
And that's the criticism, the best model we have isn't the best available. What our academics know isn't the best available predictive analytics. You simply can't built a complex ML model in C in the same way you could in Python and you wouldn't be able to train it properly without multi threading, it would take far too long.
No amount of modelling can compensate for the unknown factors. The errors in predictions, and wide ranges of estimates are due to incomplete data and uncertain assumptions.
Trafalgar have just slapped a NC poll up on their site.
And you are never going to believe it, it has exactly the same demographics as the one from September.
Outstanding consistency
Likely they've got some techie who takes txt & robo push-button responses to push-polling questions (where the REAL point is message NOT the response) then re-configures results based on demographics obtained from source OTHER than the actual texts & calls.
Then Cahaly pumps it out as another Trafalgar Group "poll".
This is all supposition - but it DOES fit the known facts (including TG's own vague "explanation" re: methodology) AND is certainly right in the Lee Atwater - Karl Rove (remember him?) wheelhouse.
Nah, the polls appear g on their website are just made up. The 3 that appeared today and caused a stir are genuine-but-crap. Notice
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
Not really, it's been very disorganised and people have been proved to be irresponsible and not isolating after testing positive and the state has been left wanting in many areas. Any thoughts of exceptionalism should have been washed away by the virus response both from the state and the people.
The NHS, academic and pharmaceutical contribution is world class.
Lots of questions over the academic response over this, especially early on. Many so called experts not leveraging standard modern tools. Much that everybody laughed at the use of Excel for of the testing data, we saw with the likes of Ferguson model is was a similar shit show of coding / obselete programming language and paradigms.
We are still seeing it now with the model for how effective a circuit breaker might be providing nonsensical results.
The obsolete programming language was C, used for linux and android and therefore probably the most important and most commonly deployed language. (Obscure namedrop: I was once stuck in a lift with that there Dennis Ritchie.) I've not been following modelgate but gather some Microsofties had a go at porting Ferguson's code to C++ for no readily apparent reason except that is what programmers like to do: reimplement the wheel in a new language.
When all this is over, we might reflect that Dominic Cummings and pb's own @SeanT were right about one thing. Britain does need to invest more in research.
C is most definitely not an obsolete language it is still widely used for good reasons. That is however not to say its necessarily the best to use for those models. I can't comment on that because I haven't seen the models but C is I believe still one of the top ten languages and is still the best to choose often for embedded purposes. Like everything it is the right tool for the right purpose.
I would have more concern if the models were done in something live javascript than c.
John Carmack looked at the code and said it looked fine.
That is more than good enough for me.
Me too. The point I was making was merely that C is an old language but that doesn't make it obsolete. I wasn't arguing it was the right choice for this however I said I didn't know as hadn't seen the model.
Half the computer world still runs on C, apache web servers, linux, most embedded stuff etc.
Half of the computing world isn't trying to predict a viral replication model within a modelled real world scenario. It is what ML is made to do, using Python which has a whole host of ML libraries and can use GPU acceleration would be the correct way of doing it. I couldn't imagine anyone in my industry creating a predictive model in C and banking is at the forefront of ML, C is just old fashioned and unsuitable for using Tensorflow or PyTorch either of which are extensively used for deep learning which is what a problem like this needs.
Sigh does no one read what you write here anymore. I didn't argue c was the correct tool I merely asserted it wasn't an obsolete language. I said the right tool for the right job. I suspect in the case of covid however the model itself is the flaw rather than the language it is written in.
Further as to machine learning personally that is over hyped and unproven. We have a huge amount of examples of machine learning where what the machine learnt isn't what people thought they had taught it. Facial recognition, predictive crime etc being the two most egregious but lots of others out there.
I think the point being made is that him using C shows how outdated his approach is.
ML is great for these kinds of problems because people are predictably unpredictable. Predicting a person's behaviour is impossible, predicting population behaviour isn't and it's one of the major uses of it.
Agree that facial recognition and crime models are a pile of dogshit, but that doesn't invalidate the approach to other more suitable problems. Viral replication is probably one of the most suitable problems for ML as it's a multivariate analysis and we have a huge amount of data available now. Another good area is medical outcome modelling, a friend of mine works for Babylon and they've been doing some incredible work on it.
Don't mistake me I am not saying machine learning doesnt have a place. Sadly like much in the it world it almost feels like its becoming the silver bullet.....whatever the problem machine learning is the answer or block chain
I'm not so sure, I've implemented ML in a few areas of my work and it's been very helpful, but I agree that it isn't a silver bullet for every scenario. It's definitely well ahead of Blockchain in real world use, I mean in gaming DLSS is about to make native resolution obsolete and that's built on ML. Blockchain is vaporware, ML definitely isn't.
I think the issue with ML is a bit like some of the government announcements, overpromising. People seem to want to claim it is the magic bullet and will solve problems to 99.9% accuracy. When it is better to say, no, but using modern ML approaches can produce results which are superior to older techniques.
e.g. For computer vision, 10-15 years ago, everybody tried to manually come up with some set of descriptors to find particular objects (or parts of objects in a scene). That approach never produced robust or generalizable solutions.
Now we have models that have learned to identify objects with a much improved level of accuracy. They can't do it 100%, they can be tricked, but in comparison to state of the art from 15 years ago, it is chalk and cheese, where we are at a stage that they models can be built into real world applications (with the knowledge they won't work 100% of the time).
Now if you want to debate if some variant of a CNN, ResNet or GAN is THE solution, now that is definitely where there is huge over-promising. They aren't, but they have shown that we can harness the incredible power of modern CPU / GPUs to process enormous amounts of data to learn a model of a problem using latent variants, which humans are incapable of doing.
As an aside the interesting side of ml that rarely gets mentioned widely is people finding ways to fool machine learning especially around computer vision. I have noted stickers and tshirts and other clothing already being sold which are claimed to confuse it.
Within the ML community it certainly gets lots of attention. Not just fooling a discriminator, but within things like GANs, the better you make the detector, the better the generator will become, thus in simple terms, the scary possibilities of Deep Fakes, machines will become better at producing them as we become better at detecting them.
Now there are some serious shortcomings of the current state of the art GANs, but somebody will come up with a newer spin on this (or a similar idea) and the concerns still stand.
It is seriously worrying right across Europe just now
I'm not really sure why the big Western European nations (excluding Germany) have all been about as bad as one another.
Because they are culturally, geographically, economically and demographically very similar?
Because, given that the UK had pretty good weather for a September, the onset of autumn is driving everyone indoors at around the same time. It really is much more dangerous mixing with strangers indoors.
I'm not so sure, I've implemented ML in a few areas of my work and it's been very helpful, but I agree that it isn't a silver bullet for every scenario. It's definitely well ahead of Blockchain in real world use, I mean in gaming DLSS is about to make native resolution obsolete and that's built on ML. Blockchain is vaporware, ML definitely isn't.
As with a lot of things the best solution is the tool you know and have to hand which was why the original prediction was written in C (and that still just a gout made sense when it was written).
And that's the criticism, the best model we have isn't the best available. What our academics know isn't the best available predictive analytics. You simply can't built a complex ML model in C in the same way you could in Python and you wouldn't be able to train it properly without multi threading, it would take far too long.
No amount of modelling can compensate for the unknown factors. The errors in predictions, and wide ranges of estimates are due to incomplete data and uncertain assumptions.
Quite. Modelling for covid is like modelling for human happiness, and trying to work out how happy a particular human will be - on a scale of 1 to 100 - in the next five, ten, fifteen years.
There are so many unknowables any attempt at precision is going to be a joke. The best you can do is input everything you know - education, class, age, wealth, health, marital status, race - and then say a prayer and press Ctrl + Prt
That is after an intense, prolonged and controversial lockdown (still ongoing in Melbourne)
We all know you can crush R down below 1 if you cease all human interaction. That’s accepted. The question then becomes: is this victory worth the economic pain? Especially if, after lockdown ends, the virus just bounces back anyhow
That is after an intense, prolonged and controversial lockdown (still ongoing in Melbourne)
We all know you can crush R down below 1 if you cease all human interaction. That’s accepted. The question then becomes: is this victory worth the economic pain? Especially if, after lockdown ends, the virus just bounces back anyhow
Here's the money quote: The poll, conducted by Scott Rasmussen,
Tie is for what the website calls Baseline Model; they also give Strong Democratic and Strong Republican turnout models, which naturally give results that show a margin of a few points either way.
SO one question is, what is more likely scenario?
My own guess is that turnout this year in Big Sky Country will tilt somewhat - not hugely, but definitely - toward Democrats.
Trafalgar have just slapped a NC poll up on their site.
And you are never going to believe it, it has exactly the same demographics as the one from September.
Outstanding consistency
Likely they've got some techie who takes txt & robo push-button responses to push-polling questions (where the REAL point is message NOT the response) then re-configures results based on demographics obtained from source OTHER than the actual texts & calls.
Then Cahaly pumps it out as another Trafalgar Group "poll".
This is all supposition - but it DOES fit the known facts (including TG's own vague "explanation" re: methodology) AND is certainly right in the Lee Atwater - Karl Rove (remember him?) wheelhouse.
Nah, the polls appear g on their website are just made up. The 3 that appeared today and caused a stir are genuine-but-crap. Notice
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
Not really, it's been very disorganised and people have been proved to be irresponsible and not isolating after testing positive and the state has been left wanting in many areas. Any thoughts of exceptionalism should have been washed away by the virus response both from the state and the people.
The NHS, academic and pharmaceutical contribution is world class.
Lots of questions over the academic response over this, especially early on. Many so called experts not leveraging standard modern tools. Much that everybody laughed at the use of Excel for of the testing data, we saw with the likes of Ferguson model is was a similar shit show of coding / obselete programming language and paradigms.
We are still seeing it now with the model for how effective a circuit breaker might be providing nonsensical results.
The obsolete programming language was C, used for linux and android and therefore probably the most important and most commonly deployed language. (Obscure namedrop: I was once stuck in a lift with that there Dennis Ritchie.) I've not been following modelgate but gather some Microsofties had a go at porting Ferguson's code to C++ for no readily apparent reason except that is what programmers like to do: reimplement the wheel in a new language.
When all this is over, we might reflect that Dominic Cummings and pb's own @SeanT were right about one thing. Britain does need to invest more in research.
C is most definitely not an obsolete language it is still widely used for good reasons. That is however not to say its necessarily the best to use for those models. I can't comment on that because I haven't seen the models but C is I believe still one of the top ten languages and is still the best to choose often for embedded purposes. Like everything it is the right tool for the right purpose.
I would have more concern if the models were done in something live javascript than c.
John Carmack looked at the code and said it looked fine.
That is more than good enough for me.
Me too. The point I was making was merely that C is an old language but that doesn't make it obsolete. I wasn't arguing it was the right choice for this however I said I didn't know as hadn't seen the model.
Half the computer world still runs on C, apache web servers, linux, most embedded stuff etc.
Half of the computing world isn't trying to predict a viral replication model within a modelled real world scenario. It is what ML is made to do, using Python which has a whole host of ML libraries and can use GPU acceleration would be the correct way of doing it. I couldn't imagine anyone in my industry creating a predictive model in C and banking is at the forefront of ML, C is just old fashioned and unsuitable for using Tensorflow or PyTorch either of which are extensively used for deep learning which is what a problem like this needs.
Sigh does no one read what you write here anymore. I didn't argue c was the correct tool I merely asserted it wasn't an obsolete language. I said the right tool for the right job. I suspect in the case of covid however the model itself is the flaw rather than the language it is written in.
Further as to machine learning personally that is over hyped and unproven. We have a huge amount of examples of machine learning where what the machine learnt isn't what people thought they had taught it. Facial recognition, predictive crime etc being the two most egregious but lots of others out there.
I think the point being made is that him using C shows how outdated his approach is.
ML is great for these kinds of problems because people are predictably unpredictable. Predicting a person's behaviour is impossible, predicting population behaviour isn't and it's one of the major uses of it.
Agree that facial recognition and crime models are a pile of dogshit, but that doesn't invalidate the approach to other more suitable problems. Viral replication is probably one of the most suitable problems for ML as it's a multivariate analysis and we have a huge amount of data available now. Another good area is medical outcome modelling, a friend of mine works for Babylon and they've been doing some incredible work on it.
Don't mistake me I am not saying machine learning doesnt have a place. Sadly like much in the it world it almost feels like its becoming the silver bullet.....whatever the problem machine learning is the answer or block chain
I'm not so sure, I've implemented ML in a few areas of my work and it's been very helpful, but I agree that it isn't a silver bullet for every scenario. It's definitely well ahead of Blockchain in real world use, I mean in gaming DLSS is about to make native resolution obsolete and that's built on ML. Blockchain is vaporware, ML definitely isn't.
By machine learning, are you talking about neural networks, or genetic algorithms? The former require tons of historical data, so are not so useful if you're dealing with inherently unknowable systems or situations where you either have no data or you only have low confidence in what you have. I think, for example, they will axiomatically fail to predict black swans or emergent properties that are not in the historical data used to train them.
For genetic algorithms, I am not sure how evolved solutions in silico will necessarily be better than those evolved through acquisition of genuine expertise.
I am not an expert in this area, and so would welcome correction and enlightenment if I am wrong.
PS These comments relate to machine learning in complex adaptive systems - I realize that it can be very effective in better defined, deterministic systems
Neither a Deep Neural Net or a genetic algorithm is really the best approach. Something like Gaussian Processes based model is (and no that doesn't mean just fit a gaussian curve though some data).
What ever was the best language to build the 'sim' virus model (which incidentally was written for an influenza type virus rather than a SARs-type) the key issue, imho, is the it has failed repeatedly to be anywhere near the turned out reality.
And as far as I am aware no changes have been made to the model based on what has actually happened in the real world (happy to be corrected on this).
The Covid Modelling Problem is like a ball rolling down a knife-edge ridge, between two deep, deep valleys.
The ball might roll down the the ridge the whole way, but way more likely, it will end up in the valley on on one side of the ridge, or the other. Small differences in starting position, velocity and angle make a huge difference as to which valley you drop into.
In one of the valleys, you have suppressed the virus, but destroyed your economy. In the other valley, you have lost complete control of the virus & the pandemic is raging.
There is only a tiny bit of middle ground on the ridge, so it looks impossible to get the ball down the ridge without it dropping in the valley.
Can we do it?
What we have to do is use human behaviour to get the ball safely down the ridge. When the epidemic gets out of control, humans must do better social distancing, so the ball can stay on track & not drop into one valley.
When the epidemic starts to fade, humans must loosen up to keep the ball on the ridge & not drop into the other valley.
It is is almost impossible to model this because it needs maths+psychology+economics. There are just huge feedback loops, as the system changes, so does peoples' behaviour.
I'm not so sure, I've implemented ML in a few areas of my work and it's been very helpful, but I agree that it isn't a silver bullet for every scenario. It's definitely well ahead of Blockchain in real world use, I mean in gaming DLSS is about to make native resolution obsolete and that's built on ML. Blockchain is vaporware, ML definitely isn't.
As with a lot of things the best solution is the tool you know and have to hand which was why the original prediction was written in C (and that still just a gout made sense when it was written).
And that's the criticism, the best model we have isn't the best available. What our academics know isn't the best available predictive analytics. You simply can't built a complex ML model in C in the same way you could in Python and you wouldn't be able to train it properly without multi threading, it would take far too long.
No amount of modelling can compensate for the unknown factors. The errors in predictions, and wide ranges of estimates are due to incomplete data and uncertain assumptions.
Firstly the criticism is that they are using inferior modelling approaches and they have produced totally unrealistic results, which suggests a problem with their model.
And secondly, actually you aren't 100% correct. The whole point of modern ML is to find and provide estimates of latent variables. We now have 6 months of data from all around the world. Yes it is noisy, yes every country has done something different, but modern ML techniques are designed to take exactly this kind of data and tease out factors which aren't known / humans are incapable of observing.
Furthermore, modern techniques are able to not only produce "error bars", but probabilistic estimates. In simplest terms, when the modellers said well number of deaths saved will be between 800 and 107k, not only was this range far too wide, they were unable to produce how different probabilistic interpretations of the different proportions of that range.
Trafalgar have just slapped a NC poll up on their site.
And you are never going to believe it, it has exactly the same demographics as the one from September.
Outstanding consistency
Likely they've got some techie who takes txt & robo push-button responses to push-polling questions (where the REAL point is message NOT the response) then re-configures results based on demographics obtained from source OTHER than the actual texts & calls.
Then Cahaly pumps it out as another Trafalgar Group "poll".
This is all supposition - but it DOES fit the known facts (including TG's own vague "explanation" re: methodology) AND is certainly right in the Lee Atwater - Karl Rove (remember him?) wheelhouse.
Nah, the polls appear g on their website are just made up. The 3 that appeared today and caused a stir are genuine-but-crap. Notice
It is seriously worrying right across Europe just now
Tbf, this is us in two weeks unless the government gets a grip on self isolation and testing failures.
An honest question
Is anyone testing in Europe any better than the UK or Germany
Not particularly, I'd say that France, Spain, Netherlands and Belgium testing is worse than our system, Germany and Italy probably marginally better.
Ultimately every country suffers from the same issue of people not isolating after testing positive. Nations who have effectively defeated the virus have very good isolation measures (Australia, NZ) or people who are responsible and isolate properly without needing special measures (SK, Japan).
We seem to delight in attacking our own system which I understand is near a 400,000 a day capacity and is not far off Germany.
And of course the countries who have best dealt with this are all in the Southern Hemisphere and in New Zealand case 1500 miles from nearest land in Australia and comprises two islands with wide open spaces
The British response to Corona has much to be proud of, with only the government letting the side down.
Not really, it's been very disorganised and people have been proved to be irresponsible and not isolating after testing positive and the state has been left wanting in many areas. Any thoughts of exceptionalism should have been washed away by the virus response both from the state and the people.
The NHS, academic and pharmaceutical contribution is world class.
Lots of questions over the academic response over this, especially early on. Many so called experts not leveraging standard modern tools. Much that everybody laughed at the use of Excel for of the testing data, we saw with the likes of Ferguson model is was a similar shit show of coding / obselete programming language and paradigms.
We are still seeing it now with the model for how effective a circuit breaker might be providing nonsensical results.
The obsolete programming language was C, used for linux and android and therefore probably the most important and most commonly deployed language. (Obscure namedrop: I was once stuck in a lift with that there Dennis Ritchie.) I've not been following modelgate but gather some Microsofties had a go at porting Ferguson's code to C++ for no readily apparent reason except that is what programmers like to do: reimplement the wheel in a new language.
When all this is over, we might reflect that Dominic Cummings and pb's own @SeanT were right about one thing. Britain does need to invest more in research.
C is most definitely not an obsolete language it is still widely used for good reasons. That is however not to say its necessarily the best to use for those models. I can't comment on that because I haven't seen the models but C is I believe still one of the top ten languages and is still the best to choose often for embedded purposes. Like everything it is the right tool for the right purpose.
I would have more concern if the models were done in something live javascript than c.
John Carmack looked at the code and said it looked fine.
That is more than good enough for me.
Me too. The point I was making was merely that C is an old language but that doesn't make it obsolete. I wasn't arguing it was the right choice for this however I said I didn't know as hadn't seen the model.
Half the computer world still runs on C, apache web servers, linux, most embedded stuff etc.
Half of the computing world isn't trying to predict a viral replication model within a modelled real world scenario. It is what ML is made to do, using Python which has a whole host of ML libraries and can use GPU acceleration would be the correct way of doing it. I couldn't imagine anyone in my industry creating a predictive model in C and banking is at the forefront of ML, C is just old fashioned and unsuitable for using Tensorflow or PyTorch either of which are extensively used for deep learning which is what a problem like this needs.
Sigh does no one read what you write here anymore. I didn't argue c was the correct tool I merely asserted it wasn't an obsolete language. I said the right tool for the right job. I suspect in the case of covid however the model itself is the flaw rather than the language it is written in.
Further as to machine learning personally that is over hyped and unproven. We have a huge amount of examples of machine learning where what the machine learnt isn't what people thought they had taught it. Facial recognition, predictive crime etc being the two most egregious but lots of others out there.
I think the point being made is that him using C shows how outdated his approach is.
ML is great for these kinds of problems because people are predictably unpredictable. Predicting a person's behaviour is impossible, predicting population behaviour isn't and it's one of the major uses of it.
Agree that facial recognition and crime models are a pile of dogshit, but that doesn't invalidate the approach to other more suitable problems. Viral replication is probably one of the most suitable problems for ML as it's a multivariate analysis and we have a huge amount of data available now. Another good area is medical outcome modelling, a friend of mine works for Babylon and they've been doing some incredible work on it.
Don't mistake me I am not saying machine learning doesnt have a place. Sadly like much in the it world it almost feels like its becoming the silver bullet.....whatever the problem machine learning is the answer or block chain
I'm not so sure, I've implemented ML in a few areas of my work and it's been very helpful, but I agree that it isn't a silver bullet for every scenario. It's definitely well ahead of Blockchain in real world use, I mean in gaming DLSS is about to make native resolution obsolete and that's built on ML. Blockchain is vaporware, ML definitely isn't.
By machine learning, are you talking about neural networks, or genetic algorithms? The former require tons of historical data, so are not so useful if you're dealing with inherently unknowable systems or situations where you either have no data or you only have low confidence in what you have. I think, for example, they will axiomatically fail to predict black swans or emergent properties that are not in the historical data used to train them.
For genetic algorithms, I am not sure how evolved solutions in silico will necessarily be better than those evolved through acquisition of genuine expertise.
I am not an expert in this area, and so would welcome correction and enlightenment if I am wrong.
PS These comments relate to machine learning in complex adaptive systems - I realize that it can be very effective in better defined, deterministic systems
Neither a Deep Neural Net or a genetic algorithm is really the best approach. Something like Gaussian Processes based model is (and no that doesn't mean just fit a gaussian curve though some data).
What ever was the best language to build the 'sim' virus model (which incidentally was written for an influenza type virus rather than a SARs-type) the key issue, imho, is the it has failed repeatedly to be anywhere near the turned out reality.
And as far as I am aware no changes have been made to the model based on what has actually happened in the real world (happy to be corrected on this).
Well this was part of the criticism. His model was made using outdated set of tools, which makes it clear he hasn't updated it and adapted to both flaws of it when it previously hit the real world (i.e. the swing flu), but neither has he kept up with the state of the art mathematical modelling techniques.
The UK prime minister and the home secretary are accused of endangering the personal safety of lawyers through their abusive attacks on the profession and should apologise, more than 800 former judges and senior legal figures have said in a letter sent to the Guardian.
Boris Johnson and Priti Patel are additionally accused in the letter of displaying “hostility” towards lawyers, undermining the rule of law and effectively risking the lives of those working in the justice system.
The signatories include three former justices of the UK supreme court, five retired appeal court judges, three former high court judges, the lawyer heads of four Oxford University colleges, more than 80 QCs, 69 law professors from leading English universities, the directors of Liberty and Justice, as well as hundreds of law firm partners, barristers and solicitors.
Comments
If a distribution has a fat tail, most of the information is away from the centre of the distribution. For a fat tail, it is not possible or sensible to make a point forecast. All that is possible is a distributional forecast.
I don’t think Neil Ferguson's models are flawed because of poor data modelling practice. They have flaws, but the flaws are inevitable — wrong assumptions and high sensitivity are unavoidable in modelling the complex effects of something that has never happened before on the human population.
The sad truth is I’m afraid: point forecasts are close to useless and distributional forecasts are really hard.
I am much more of a COVID empiricist.
We can see what works by looking around the world. And we should put our faith in that, and not in modelling which does not set out to do what you want it to.
Ferguson has received a lot of criticism but much of it simply misunderstands what is even possible with this kind of modelling.
If you want results fast go multithreaded and parallel. If you dont mind if it takes a few hours you are more likely to get reliable results with a single threaded app as you havent added possible error sources by hoping no one has forgotton to use a mutex/critical section/semaphore etc where they should have. Use it when you need to but be aware of the drawbacks
That isn't how being a world leader is suppose to be. You should be constantly improving with the times.
The past 10 years has seen incredible progress in machine learning. He hasn't kept up with the times.
And this was the best we had. The original suggestion was academia were brilliant in response to this pandemic, it was just the government who were poorly prepared and crap.
More people - and certainly more people who aren’t already very elderly - are dying of cancer and heart failure each week, and yet it would never enter our minds to incur this level of economic damage to drive down those deaths.
Basically, no one really knows. There are so many imponderables it is beyond the wit of man to know the future of this virus.
However the University of Washington Covid Model, which used to be laughably poor, now looks increasingly prescient, as they keep pumping in more data.
FWIW by Feb 2021 they predict 150,000 dead in the UK, and 385,000 dead in the USA. Grisly. Let’s hope they are wrong
https://covid19.healthdata.org/united-states-of-america
In US authorities almost always blood alcohol percents (determined by blood sample or more commonly by breathilizer) to determine if drivers are over the limit. Also field sobriety tests pretty common.
That's IF you get pulled over for erratic driving OR are involved in a accident and appear to be impaired.
When I have to celebrate electoral success 8 time zones away I guess I'm struggling.
Keep in mind that drunk driving is a VERY serious problem across the USA where even today car ownership (and culture) is a BIGGER deal than in Europe (including obscure offshore islets),
It is highly covid cautious in its check-in procedures, elevator protocols, room cleaning, etc.
Yet the restaurant is absolutely packed and rocking with maskless people quaffing wine and expelling aerosols all over the shop. There are lots of bars and pizzerias and sushi bars all around, and they are similarly busy. It’s very upbeat.... but I just don’t see how we avoid following Italy, France, Spain, etc, into a pretty severe second lockdown and early curfews. Britain is not exceptionally immune. Just late to the party. Again.
On the upside Bristol - a city I barely knew - turns out to be highly beguiling. A hidden gem. When the Wuhan Flu is over I might return to buy a plague-vacated Georgian house for £10.
To get to those level of figures the hospitals will be flooded with patients.
Tesco Pengam Green will be like Harrod's January sale as the doors open at 9 am on New Year's day.
For quite a while one of the best COVID-19 models was one guy from MIT applying machine learning to an off-the-shelf model. He was consitently beating all sorts of fancy national labs and universities, many of which were so crap at modelling that they could not even beat the baseline of repeating the previous measured change. To be fair modelling has improved dramatically over the last six months or so, but I don't think it's unreasonable to state that the bulk of epidemiological modelling when the pandemic hit was a load of rubbish.
Further as to machine learning personally that is over hyped and unproven. We have a huge amount of examples of machine learning where what the machine learnt isn't what people thought they had taught it. Facial recognition, predictive crime etc being the two most egregious but lots of others out there.
https://twitter.com/PaulNuki/status/1320424566691373056?s=20
But quietly and relentlessly they have been honing it, and now it appears rather on point.
385,000 dead in the USA by Feb 1 2021 is actually optimistic compared to some other forecasts.
Furthermore it isnt just a C vs python thing. The world of mathematical modelling has been revolutionized in the past 10 years (and not just deep neural nets).
Instead the world leading virus modeller is still using a model from 10+ years ago.
Of course, if would be best to have both.
ML is great for these kinds of problems because people are predictably unpredictable. Predicting a person's behaviour is impossible, predicting population behaviour isn't and it's one of the major uses of it.
Agree that facial recognition and crime models are a pile of dogshit, but that doesn't invalidate the approach to other more suitable problems. Viral replication is probably one of the most suitable problems for ML as it's a multivariate analysis and we have a huge amount of data available now. Another good area is medical outcome modelling, a friend of mine works for Babylon and they've been doing some incredible work on it.
Bristol was the favorite British city of an old friend of mine. She only visited it once but had a really good time and really loved the place. Which was HUGELY important to early British colonialism and mercantilism, and to the settlement and development of (what became) the United States.
https://twitter.com/KFaulders/status/1320460933911748610?s=19
Indeed is IS a scandal, what Trumpskyites have been doing here. Sadly for them, turns out to be a boomerang.
The interesting bit would be getting the ML "thinkers" to have a look at the whole area - looks like that wasn't in Ferguson's wheelhouse.
Sadly he's stopped running his model (given the amount of work he's done I should think he needs a rest), but there is a hell of a lot of good stuff to read on his site. It's well worth looking at his ranking for the various US models, a lot of very prominent organisations have been pumping out absolute rubbish for months on end.
Check out the first big colour-coded table for comparisons on the page below.
https://covid19-projections.com/about/
Getting closer.
e.g. For computer vision, 10-15 years ago, everybody tried to manually come up with some set of descriptors to find particular objects (or parts of objects in a scene). That approach never produced robust or generalizable solutions.
Now we have models that have learned to identify objects with a much improved level of accuracy. They can't do it 100%, they can be tricked, but in comparison to state of the art from 15 years ago, it is chalk and cheese, where we are at a stage that they models can be built into real world applications (with the knowledge they won't work 100% of the time).
Now if you want to debate if some variant of a CNN, ResNet or GAN is THE solution, now that is definitely where there is huge over-promising. They aren't, but they have shown that we can harness the incredible power of modern CPU / GPUs to process enormous amounts of data to learn a model of a problem using latent variants, which humans are incapable of doing.
https://twitter.com/Keir_Starmer/status/1320474208917479425
Beyond that, is trying to establish some basis of relationship with (increasingly likely) Biden administration.
What a coincidence!
Which btw, should tell you everything you need to know about the hypocrisy of the Democrats talking about Trump and Russian interference.
'I won't need - I don't think - to buy clothing over this two weeks and I think many many people in Wales will be in that poition to. For me it won't be essential but I recognise some people for entirely unexpected reasons which they could not have foreseen will need to buy items'
So laws in Wales are being made by a first minister who does not think
For genetic algorithms, I am not sure how evolved solutions in silico will necessarily be better than those evolved through acquisition of genuine expertise.
I am not an expert in this area, and so would welcome correction and enlightenment if I am wrong.
PS These comments relate to machine learning in complex adaptive systems - I realize that it can be very effective in better defined, deterministic systems
Mr Faure, who had inherited power from his predecessor four years ago, was unable to distance his party's campaign from mounting evidence of past political murders, torture and corruption when Seychelles was still a one-party state.
https://www.bbc.co.uk/news/world-africa-54681360
the FL, AZ, MI polls are push polls with small sample sizes and HILARIOUS demographic breakdowns.
Just like their 2016 polling they have Trump winning the 18-24 year old demographic.
https://electproject.github.io/Early-Vote-2020G/index.html
Now there are some serious shortcomings of the current state of the art GANs, but somebody will come up with a newer spin on this (or a similar idea) and the concerns still stand.
http://politicaliq.com/2020/10/25/mt-us-congress-williamsd-47-rosendaler-47/
Here's the money quote: The poll, conducted by Scott Rasmussen,
https://twitter.com/BNODesk/status/1320486138885394434?s=19
There are so many unknowables any attempt at precision is going to be a joke. The best you can do is input everything you know - education, class, age, wealth, health, marital status, race - and then say a prayer and press Ctrl + Prt
Great, and a vast improvement on Corbyn who would probably be all mealy mouthed.
Trump only has his family and the lawyer with his hand down his trousers left.
We all know you can crush R down below 1 if you cease all human interaction. That’s accepted. The question then becomes: is this victory worth the economic pain? Especially if, after lockdown ends, the virus just bounces back anyhow
SO one question is, what is more likely scenario?
My own guess is that turnout this year in Big Sky Country will tilt somewhat - not hugely, but definitely - toward Democrats.
And as far as I am aware no changes have been made to the model based on what has actually happened in the real world (happy to be corrected on this).
The ball might roll down the the ridge the whole way, but way more likely, it will end up in the valley on on one side of the ridge, or the other. Small differences in starting position, velocity and angle make a huge difference as to which valley you drop into.
In one of the valleys, you have suppressed the virus, but destroyed your economy. In the other valley, you have lost complete control of the virus & the pandemic is raging.
There is only a tiny bit of middle ground on the ridge, so it looks impossible to get the ball down the ridge without it dropping in the valley.
Can we do it?
What we have to do is use human behaviour to get the ball safely down the ridge. When the epidemic gets out of control, humans must do better social distancing, so the ball can stay on track & not drop into one valley.
When the epidemic starts to fade, humans must loosen up to keep the ball on the ridge & not drop into the other valley.
It is is almost impossible to model this because it needs maths+psychology+economics. There are just huge feedback loops, as the system changes, so does peoples' behaviour.
And secondly, actually you aren't 100% correct. The whole point of modern ML is to find and provide estimates of latent variables. We now have 6 months of data from all around the world. Yes it is noisy, yes every country has done something different, but modern ML techniques are designed to take exactly this kind of data and tease out factors which aren't known / humans are incapable of observing.
Furthermore, modern techniques are able to not only produce "error bars", but probabilistic estimates. In simplest terms, when the modellers said well number of deaths saved will be between 800 and 107k, not only was this range far too wide, they were unable to produce how different probabilistic interpretations of the different proportions of that range.
I just can't find a source for 2016 congressional polls to cross check.
Boris Johnson and Priti Patel are additionally accused in the letter of displaying “hostility” towards lawyers, undermining the rule of law and effectively risking the lives of those working in the justice system.
The signatories include three former justices of the UK supreme court, five retired appeal court judges, three former high court judges, the lawyer heads of four Oxford University colleges, more than 80 QCs, 69 law professors from leading English universities, the directors of Liberty and Justice, as well as hundreds of law firm partners, barristers and solicitors.
Once upon a time, they’d all have been Tories.