|
Huh, looks like I'm back teaching in a classroom sometime next week or the week after.
|
# ? May 7, 2020 22:27 |
|
|
# ? Jun 12, 2024 15:27 |
|
Nu kan du sova lugnt PederP, den modell Imperial college använder tycks vara katastrofalt dåligt kodad, . Sverige kommer alltså inte utplånas av den öppna strategin. https://github.com/mrc-ide/covid-sim/issues/165?fbclid=IwAR03brNit45hD14nkm_OpXqa_Lq0a-_YlsWE4v02qQmA56mWU0V3Gl3xAa8
|
# ? May 8, 2020 04:42 |
|
https://twitter.com/KGuw/status/1258492115941810176?s=19
|
# ? May 8, 2020 06:24 |
|
Potrzebie posted:Nu kan du sova lugnt PederP, den modell Imperial college använder tycks vara katastrofalt dåligt kodad, Oh please, I sleep just fine. Just because I worry about other people and the society I live in doesn't mean I'm a quivering wreck of anxiety. I don't expect you'll be "utplånat", which I assume means wiped out (my Swedish is pretty bad), but I think you are, and will have, tons of unnecessary deaths. The revelation that most scientists write incredibly lovely code and that unit testing fundamentalists are being dumb fanatics in public again, will not bring back the dead who could have been saved. But if it makes you sleep better at night to believe your authorities did the right thing, feel free to gloat about a deadly pandemic on this dead gay comedy forum.
|
# ? May 8, 2020 06:29 |
|
murphyslaw posted:Huh, looks like I'm back teaching in a classroom sometime next week or the week after. Yeah same here. From Tuesday it seems, though I'm awaiting information from the adminstration at my school for more precise confirmation and details.
|
# ? May 8, 2020 06:41 |
|
Potrzebie posted:Nu kan du sova lugnt PederP, den modell Imperial college använder tycks vara katastrofalt dåligt kodad,
|
# ? May 8, 2020 07:31 |
|
A Buttery Pastry posted:Aha, I told you it was suspiciously opaque! Your assumption that this means good things for Sweden doesn't follow however, the Imperial college model might be lowballing casualties for all we know. Touché! But nothing at present indicates that it is even half as bad as the model predicts.
|
# ? May 8, 2020 09:37 |
|
A Buttery Pastry posted:Aha, I told you it was suspiciously opaque! Your assumption that this means good things for Sweden doesn't follow however, the Imperial college model might be lowballing casualties for all we know. When applied to Sweden it predicted 40,000 deaths by early May (10,000 if they had imposed total lockdown) and 100,000 by the end of June so I'm gonna go out on a limb and say it's not lowballing and is instead poo poo.
|
# ? May 8, 2020 09:55 |
|
Maybe it is just my lack of programming experience, but I'm sort of struggling to see how the criticism of the testing regime automatically leads to the conclusion that all papers using the model should be retracted? It's certainly worth looking into if this actually impacts whether the model working as intended, but I'm not really seeing any proof of this being the case.
|
# ? May 8, 2020 10:20 |
|
military cervix posted:Maybe it is just my lack of programming experience, but I'm sort of struggling to see how the criticism of the testing regime automatically leads to the conclusion that all papers using the model should be retracted? It's certainly worth looking into if this actually impacts whether the model working as intended, but I'm not really seeing any proof of this being the case. Anyone who expects unit testing to be the norm in scientific computing is going to be disappointed. The overall code quality also tends to much lower and the code is more akin to a hacked-together script. Un-validated third party libraries are also commonly used and simply assumed to be correct. Is there a need for higher quality in this area, including better use of automated testing and verification frameworks? Absolutely. Is the FHI code of a high quality compared to international standards? Absolutely. Is automated testing the alpha and omega of software development? Absolutely not. In scientific computing I'd argue that use of formal verification frameworks is more important.
|
# ? May 8, 2020 10:26 |
|
the model had a very good fit during the early outbreaks, but seems to have expected people to be less disciplined in their distancing efforts than they were. there are going to be lots of serious people doing evaluations of the model in the time ahead, no need for amateurs to start getting in on it
|
# ? May 8, 2020 10:50 |
|
Unit testing is extremely overrated. In many projects, especially those with formalized expectations of test coverage, it simply balloons the size and cost of the project, wasting man-hours on writing tests for simple code that is highly unlikely to fail, clutters the code base and introduces it's own running costs in the form of maintenance and justification for hiring various busy-body middle managers and "testing strategy experts" that do anything but actually write code. I'm slightly more receptive to functional testing, and even more so towards integration testing, but all of these disciplines have costs and trade-offs, and they're easy to do wrong.
|
# ? May 8, 2020 10:51 |
|
Potrzebie posted:Touché! But nothing at present indicates that it is even half as bad as the model predicts. Mooey Cow posted:When applied to Sweden it predicted 40,000 deaths by early May (10,000 if they had imposed total lockdown) and 100,000 by the end of June so I'm gonna go out on a limb and say it's not lowballing and is instead poo poo. V. Illych L. posted:the model had a very good fit during the early outbreaks, but seems to have expected people to be less disciplined in their distancing efforts than they were. there are going to be lots of serious people doing evaluations of the model in the time ahead, no need for amateurs to start getting in on it
|
# ? May 8, 2020 11:03 |
|
Good thing the models employed by the Danish Health Authority are literally classified and the numbers released totally opaque, someone might notice they forgot to carry the 1 or something. Open 'er up and let God sort 'em out, I say.
|
# ? May 8, 2020 11:12 |
|
Is there data yet on the demographics of the deceased in Sweden? Like, age, how many were sick beforehand or something?
|
# ? May 8, 2020 11:14 |
|
Zzulu posted:Is there data yet on the demographics of the deceased in Sweden? Like, age, how many were sick beforehand or something? This has ages: https://experience.arcgis.com/experience/09f821667ce64bf7be6f9f87457ed9aa
|
# ? May 8, 2020 11:24 |
|
SplitSoul posted:Good thing the models employed by the Danish Health Authority are literally classified and the numbers released totally opaque, someone might notice they forgot to carry the 1 or something. Open 'er up and let God sort 'em out, I say. Look up Landspatientregisteret (LPR 3). This pandemic coincided with yet another public IT system mess. Actually getting access to the data they need is... problematic. Opening up the code might reveal just how much is based on estimates rather than actual data.
|
# ? May 8, 2020 11:27 |
|
Was that done in-house? We didn't do it.
|
# ? May 8, 2020 11:47 |
|
SplitSoul posted:Good thing the models employed by the Danish Health Authority are literally classified and the numbers released totally opaque, someone might notice they forgot to carry the 1 or something. Open 'er up and let God sort 'em out, I say. unironically i think public IT and modelling stuff should be open source unless there's an incredibly compelling reason not to do it that way like, i can see how you'd want to keep your military communications protocols under lock and key, but that's about it. open source is normally good for security
|
# ? May 8, 2020 11:52 |
|
https://www.berlingske.dk/kommentatorer/coronatal-skal-frem-i-lysetquote:Forrige uge kom nyheden om, at smittetrykket i samfundet var steget fra 0,6 til 0,9. Smittetrykket er det, der på fagsprog kaldes reproduktionsraten, og er et af nøgletallene til at guide beredskabet. Er smittetrykket 0,9, betyder det, at hver enkelt coronasmittet person i gennemsnit smitter 0,9 andre personer. Men det er os fuldstændigt uklart, hvordan man er kommet frem til et smittetryk på 0,9. Når vi selv efterprøver tilgængelige data og tabeller, stemmer tallene simpelthen ikke. Vi kan ikke få oplyst, hvordan myndighederne opgør tallet, selvom det er en af de mest essentielle oplysninger for at forstå udbruddets udvikling. Der er sikkert en god forklaring. Vi kender den bare ikke.
|
# ? May 8, 2020 11:54 |
|
no loving way
|
# ? May 8, 2020 11:56 |
|
V. Illych L. posted:unironically i think public IT and modelling stuff should be open source unless there's an incredibly compelling reason not to do it that way Whaaat? You don't think it's appropriate to pay billions for proprietary products that either don't work at all and have to be binned, or work so poorly they endanger patients by multiplying their dosages or list innocent citizens as having been charged with crimes? Clearly you're not cut out for politics, friend. As for military communications, they've been hit by simple phishing scams numerous times and have exceptionally poor password discipline. I also know of one instance where a person contracted for a network solution was handed an admin password before receiving security clearance. It boggles the mind.
|
# ? May 8, 2020 12:12 |
|
SplitSoul posted:Whaaat? You don't think it's appropriate to pay billions for proprietary products that either don't work at all and have to be binned, or work so poorly they endanger patients by multiplying their dosages or list innocent citizens as having been charged with crimes? Clearly you're not cut out for politics, friend. When I was doing my conscription service some 15 years ago, I was working radio coms in one of the most secure bunkers in Norway. I also had DB access to all the user accounts on the internal base network and while I couldn't view passwords, I could create new accounts and change access. My actual role was to man a radio. That eventually got fixed, some 3 years after me and every other IT savvy conscript pointed out how laughable it was. I imagine it'd been like that for years before my service though.
|
# ? May 8, 2020 13:06 |
|
V. Illych L. posted:unironically i think public IT and modelling stuff should be open source unless there's an incredibly compelling reason not to do it that way Robert Skov, head of the analysis group in charge of this model, has this hot take: quote:Han opfordrer ekspertgruppen til at lægge kildekoden og data frem, så andre forskere både kan forstå og bidrage til arbejdet med prognoserne. Then there's this nugget from October regarding the system supplying the data for the models: quote:Tekniske problemer med Landspatientregisteret har betydet, at ingen i øjeblikket har overblik over aktiviteterne på landets sygehuse. Last I heard (February) they were engaged in a legal slapfight with the supplier of the IT system. I wouldn't expect them to have fixed it in the middle of a pandemic. The model and the data are not classified for reasons of national security - they're being kept away from scrutiny, because the data quality is so horrendously bad that there would be an uproar. Additionally, the group seems to have be working on a brand new modified SEIR-family model that they don't want to spill the beans on before they're ready to publish it. Also, the head of SSI is currently suspended due to an investigation into corruption and illegal use of health data (including sending it to partners in the US): https://www.dr.dk/nyheder/indland/direktoer-i-statens-serum-institut-fritaget-tjeneste There's scandals and corruption lurking behind the lack of transparency.
|
# ? May 8, 2020 16:15 |
|
Holy gently caress, why do they keep giving public contracts to DXC (and formerly CSC) in Denmark. https://www.cisomag.com/software-bug-exposes-cpr-numbers-of-1-26-million-danish-citizens/ https://m.itwatch.dk/article/11439043 https://www.version2.dk/artikel/hov-hvor-blev-csc-dansk-hovedpine-slaaet-sammen-med-hps-dyre-fejltagelse-1076483 Rust Martialis fucked around with this message at 17:10 on May 8, 2020 |
# ? May 8, 2020 17:03 |
|
Rust Martialis posted:Holy gently caress, why do they keep giving public contracts to DXC in Denmark. Because the entire procurement system is rotten and broken.
|
# ? May 8, 2020 17:10 |
|
PederP posted:Anyone who expects unit testing to be the norm in scientific computing is going to be disappointed. The overall code quality also tends to much lower and the code is more akin to a hacked-together script. Un-validated third party libraries are also commonly used and simply assumed to be correct. Is there a need for higher quality in this area, including better use of automated testing and verification frameworks? Absolutely. Is the FHI code of a high quality compared to international standards? Absolutely. Is automated testing the alpha and omega of software development? Absolutely not. In scientific computing I'd argue that use of formal verification frameworks is more important. Don’t dox me. Also lol if you think this is ever going to happen. Academic scientific code is written to solve a specific problem and general applicability is a bonus rather than a feature. If you want do something not standard, prepare to either write something from scratch or write a lot of wrappers to convert between different formats. Code maintenance don’t generate grants so either code is abandoned once the grant money runs out or it is maintained by the inventor (since that is the only person who can debug the non commented code ). Some of the most used software in certain fields was written in the 70-80s using FORTRAN and nowadays I am happy if something is actually written in C++.
|
# ? May 8, 2020 17:40 |
|
Cardiac posted:Some of the most used software in certain fields was written in the 70-80s using FORTRAN and nowadays I am happy if something is actually written in C++. Why newer coders seem unable to learn FORTRAN remains a curiousity.
|
# ? May 8, 2020 17:55 |
|
The frustration is real. My Phd work has been seriously hindered by a japanese research group no longer having available grant money for their niche database. They made large updates to it about two years ago, before just replacing the whole thing by a picture of a cat without any prior warning. At least it's a pretty cute cat.
|
# ? May 8, 2020 19:15 |
|
military cervix posted:The frustration is real. My Phd work has been seriously hindered by a japanese research group no longer having available grant money for their niche database. They made large updates to it about two years ago, before just replacing the whole thing by a picture of a cat without any prior warning. At least it's a pretty cute cat. this is worse than what it's like in my field where the main database was maintained through the personal clout of literally some dude, and when he died their funding dried up immediately and moved to a hilariously unaffordable subscription model
|
# ? May 8, 2020 19:48 |
|
i've also spent non-trivial amounts of time in the last week trying to add a feature to a piece of ludicrously bloated software whose comments are deeply cryptic and also in french
|
# ? May 8, 2020 19:49 |
|
V. Illych L. posted:this is worse than what it's like in my field where the main database was maintained through the personal clout of literally some dude, and when he died their funding dried up immediately and moved to a hilariously unaffordable subscription model Oh, you didn't know? A good portion of public grants for science and tech is administrated outside any kind of law or framework, it's basically personal connections between the grants people and the scientists. No oversight. No rules. No EU regulation to speak of either, and none of the EEA economic safeguards for public expenditure apply. That turns into exactly this kind of mickey mouse bullshit and I loving hate appointees, comittee members and loving POLITICIANS ARE THE BANE OF MY loving EXISTENCE but I'm not bitter or anything no, not at all
|
# ? May 8, 2020 19:54 |
|
Cardiac posted:Don’t dox me. I know. That was my point. Corporate software developers who think their (often cargo cult-like) understanding of testing applies to every piece of code everywhere are wrong. But I do think more use of formal verification frameworks and functional programming would be a good thing. I know that the FORTRAN, C/C++ and R codebases are not going to be thrown out, but hopefully logic and functional programming will see more use for new projects. I know Scala is getting traction in some academic circles. I hate Scala, but I can appreciate the general principle.
|
# ? May 8, 2020 20:29 |
|
Programs in academia are great. In my experience most of it are written by mostly self taught people who make it for a specific task, a task that then mutates into something barely recognizable as the project evolves and the coder gains experience. An effect of this is that core features are not commented and when you ask the creator three years later how it works they glance over it and say "I... dont know". Then if the PI gets enough grant money you end up with a version 2 of the software that is functionally the same as version 1 but with a flashier interface and a new feature added on by duct tape and a prayer.
|
# ? May 8, 2020 20:44 |
|
Zudgemud posted:Programs in academia are great. In my experience most of it are written by mostly self taught people who make it for a specific task, a task that then mutates into something barely recognizable as the project evolves and the coder gains experience. An effect of this is that core features are not commented and when you ask the creator three years later how it works they glance over it and say "I... dont know". Then if the PI gets enough grant money you end up with a version 2 of the software that is functionally the same as version 1 but with a flashier interface and a new feature added on by duct tape and a prayer. Again, don’t dox me. Well, you already have..... Although I have never gotten a grant for making a program. One should note that in applied sciences programming is a tool and often handed off to the one person who is semi literate in computers. So much off the stuff is also written by a PhD or postdoc who disappears when done leaving the code behind uncommented and who likely have not much interest in maintaining it. Especially given that any reasonable programmer will go to industry for vastly better salary and working conditions. I know one AP, who is an excellent programmer in my field, who just switched to Sony once his grant ran out. The only way you can have maintainable software is by developing something so essential in your field that all articles reference you (and even that doesn’t help you if you work at NIH as another person found out when he started a private company). And don’t get me started on the licensing fees for industry for all this stuff. If something is written by accelrys or Schrodinger, prepare to be extorted. TL/DR: scientific programming is a god forsaken hellhole that is never going to change. 15 years of experience tells me this.
|
# ? May 8, 2020 21:14 |
|
i mean in fairness biology has a few solid institutions like EBI and NCBI for the most mainstream stuff, i'm sure it's worse elsewhere
|
# ? May 8, 2020 21:24 |
|
Zudgemud posted:Programs in academia are great. In my experience most of it are written by mostly self taught people who make it for a specific task, a task that then mutates into something barely recognizable as the project evolves and the coder gains experience. An effect of this is that core features are not commented and when you ask the creator three years later how it works they glance over it and say "I... dont know". Then if the PI gets enough grant money you end up with a version 2 of the software that is functionally the same as version 1 but with a flashier interface and a new feature added on by duct tape and a prayer. I remember some (associate) professors not really being able/willing to code and handing out ECTS to students who would take on the implementation part of some research project. That can lead to some pretty interesting code when the students turn out to be not just slightly overconfident, but also in a rush due to spending their time partying and gaming, instead of doing the implementation work they promised to have ready by Tuesday. It's really a weird and wonky world. Having seen code from academia, the games industry, the finance sector and generic B2B consulting/contracting work - I think many of those developers who stick to one kind of work don't appreciate just how different the culture is in other sectors. This impacts not just the work environment and the process, but the code itself. There's been some convergence during the last decade or so, due to open source and other things, but there's still an amazing variety of not just team dynamics, but also coding styles. I hope never to lose my love and appreciation for a organically grown code. Much of the great tech and software out there started out this way and then morphed into the world of corporate and enterprise software and systems, as a team grew into a company and into a corporation. I think one of the big failings of public IT is that it skips this step and just starts at the final stage. I think all public institutions and organizations should have some degree of in-house development capacity. It doesn't matter if it's a property tax office, water pollution monitoring or municipal daycare administration - if you have to go the route of procurement and external contractors for everything, the result is going to be expensive, late and/or defective. It annoys me to no end, that the public sector simply refuses to hire software engineers. Yes, it's a bit of an issue that the salaries of "line" staff is higher than middle management - but is it really worth several factors of extra cost to obscure this factor by working through greedy development companies with a small squad of high-profile consultants to keep up the charade, while milking public funds as they send actual development off to some other country to keep actual development costs to a minimum. I miss the days when it was ok to have systems written by some self-taught goofball in a not entirely appropriate programming language. It got the job done far cheaper and better than the expensive mess we have today, that just leads to one scandal after another, while pouring billions into the pockets of companies specializing in this kind of parasitic development.
|
# ? May 8, 2020 21:25 |
|
Expectations for coding ability vary wildly from field to field. My supervisor still talks warmly of the "coding work" I did during my master's, which was essentially a running a library function in a for-loop over every row in a csv-file. On a related topic, the goverernment acquisitions processes doesn't really seem to be capable of choosing functional software. At NTNU we changed to the e-learning system Blackboard a couple of years ago, which has to be the worst piece of software that I've ever seen implemented at such a large scale. It is a prime example of how massive feature creep allows a product to tick all the boxes in a procurement process, while still being absolutelely terrible to work with in practice.
|
# ? May 8, 2020 23:11 |
|
Long article on Statoil/Equinor's incredibly sloppy operations in the US in Dagens Næringsliv today. It's baffling to me that a company that's supposed to make money can be constantly leaking cash in the way they did and just lose track of tens of millions of dollars over and over, it's pretty incredible. They lost more than 200 billion NOK on the whole ordeal.
|
# ? May 9, 2020 07:41 |
|
|
# ? Jun 12, 2024 15:27 |
|
Can't code, but I've been on the other end in a few different jobs and it's always and without fail a variation on, "We got a new system that doesn't do everything the old one did, only one or two people have been properly trained in it, everyone uses the old one when the boss isn't looking even though they're not supposed to, so they don't need constant assistance to be able to do their job". I can also reveal that, as of 2016, the largest union in the country still used Windows XP due to proprietary software.
|
# ? May 9, 2020 09:14 |