Posts by ibmalone 1416 publicly visible posts • joined 6 Jul 2017
This is not actually impossible to do, for example you can feed in reams of perturbed example data and see which perturbations affect the outcome. More efficient methods might be possible for specific learning frameworks.
"it's all there, detailed across 491,840 lines of plain text."
It's a pity they didn't use .xls, this would have limited the damage to under 17 thousand.
(Also, does anyone else have the feeling Mr Dabbs has somehow arranged this as a follow-up to last week's article? A bit too neat if you ask me.)
This is particularly a problem with 'un-invigilated internet open book' versus 'you can bring in your notes' open book. It is much harder to create questions where an exact worked solution isn't out there somewhere, and at some point you want to be able to check the candidate can actually understand and apply the subject rather than simply copy someone else's solution. Additionally, student's commitment to helping each other out sometimes extends to sharing the answers with each other during the exam.
One Coursera course particularly sticks in my mind; students were to grade each-others answers after submissions for each exercise closed. You got used to noticing somewhat weird and quirky solutions, typing them into google, and finding exactly the same code on github dating to a year or two ago. Was never clear if the plagiarism reports went anywhere. (There was also the particularly special case of someone suggesting that following the marking rubric was the Nuremberg defence...)
Oh hey, just spent a bit of time helping someone out with a scrap of code I sent them last night which they couldn't get working. Turns out that copy and paste of my solution from email somehow screwed up the correct indentation while looking correct, they tried correcting it and couldn't get it right. They're not a python expert, so I'm sure it's one of our faults that this happened somehow rather than an issue with any language decisions.
The thing we're still working on (and it's for an academic environment where people justifiably need to run a whole lot of different things, some of which they are building themselves, so locking everything down is not an option), is a core console profile which sets up the preferred environment (including Python and the 'production' libraries). Then for python use conda environments (could do virtualenv instead, we happen to be using anaconda), some pre-defined and installed for tasks. The missing piece is a conda-activate-like command that will switch between base installs, not done yet but should be relatively trivial, it's just not provided by core anaconda.
For anything that needs to be run in a stable configuration it gets stuck behind wrappers that set up the environment first. Docker or other containers would work here too, we just haven't needed to do it yet.
To me it's a solution to something which isn't really a problem. The language syntax determines levels, whether by whitespace or braces, but it can't tell you what the original programmer intended. Python whitespace is more easily messed up than a brace. I write a lot of python, I don't generally have to give it much thought, but it's always there nudging your elbow a little. Sure, the errors will tell you what line has gone wrong, this is rarely a problem, the thing you want to know is what part of the flow does a line really belong to? You don't know because it's absolutely lost. Missing brace or semicolon? Learn to read the compiler errors, get compilers that produce better errors, it's not the 60's any more.
Also, it's likely you don't wish XML had Python indentation because you'd need a second monitor for most of the levels.
I was of course questioning that statement (having myself come across bugs caused by indenting mix ups, but surely no poster would ever be dishonest), and doing so by attempting to illustrate my point with a series of statements at arbitrary indentation to highlight the ambiguity it creates. The forum software decided to improve on my post by stripping the unnecessary white space, thereby making an even better point about the role of presentation in flow control.
Asserting that a mild objection, and maybe poking a little fun, is irrational hatred and prejudice doesn't make it so. It does make you look a touch irrational though.
I like python, but the white space is frustrating. It's fragile to moving code blocks or copy and paste. Making changes to code flow involves changing block indents directly and often results in re-editing lines.
White space is presentation. A decent editor can re-indent according to syntax for you, and lines can't easily slip past braces when editing, but indentation can easily change.
This means in other languages white space is not a duplication of the information braces provide, it's a visual parity check.
As for packaging, most Linux distros ship Python by default and most people that I am familiar with use their Linux distro packaging system by preference when installing packages. There's no problem with Pip from their perspective because they don't use it, and their sys admins don't want Pip used either.
This is okay, provided you only want to use libraries and programs that have been packaged by your distro. However, once you need to use other software, particularly more than one piece of other software, with different ideas about what versions it wants, things get a bit more complicated. Additionally you've got people who need to develop the things you are actually using, a handful of people who need to both use and develop things, and people who, for reasons best known to themselves, insist on using Macs.
Ironic, needing a Philips screwdriver to open a Sony product.
Maybe it was specified in the Red Book?
Only tangentially related, but it involves CDs. Around the time that x24 and x32 CD drives were newfangled and exciting, we had added a Creative-branded one to the home PC. For reasons that escape me it was living for a brief time on the kitchen table. Enter a Championship Manager CD (my brother's, recreational spreadsheets are not my thing), with the tiniest crack on the inner rim. We inspect it and decide it should be fine.
Reader, it was not fine.
On insertion the drive woke up and read the disc, okay. We then try to install, at this point the device kicks into its high speed mode with the asthmatic spin cycle sound that will still startle those who have only ever used USB sticks. There is a bang. The front cover of the tray (formerly stuck on the end of the caddy), shoots past my ear, closely followed by a fragment of now less-than-compact disc and bounces off the wall. After opening up the drive and removing the remaining pieces of the now vastly improved championship manager we reattached the front cover and found the thing still worked. It continued in service in my parents' house until replaced by a DVD drive, and was always particularly helicopter like when it started up.
Feast your eyes: https://cpc.farnell.com/pro-elec/9798/socket-twin-switched-dp/dp/PL09216
A UK plug with bare wire coming out is unlikely. I don't think sockets are required to be switched, however it's useful if you need to turn something off in a hurry. When things start smoking pulling plugs is good for ratcheting up the tension, just flicking all switches off is much less dramatic. It's also somewhat tidier to be able to leave something plugged in but switched off. End of the world either way? Not really.
I've got ZXing installed and not seen this, it's still on the play store and was last updated in 2018, so my guess is people just searching for Barcode Reader to leave bad reviews and not noticing it's a different publisher to the one they had (particularly as the malicious one has been removed).
That of course gets us into the very grey area of how much you should write yourself and how much you should depend on others to do better than you can, and there are definitely places to draw the line and times to move it. Reimplementing things though can also introduce new issues, so in the context of what may be a validated and tested piece of software, moving it to new and new language and libraries that claim to do the same thing as the old ones doesn't necessarily gain anything.
My last comment was really directed against whoever downvoted simply asking for clarification on this, so if that's not you then don't take it personally. I'm always happy to learn, but I do remain a bit sceptical; I've not seen anyone discuss what the software in question actually does. Without knowing that then it is surely a bit difficult to say whether they've chosen an appropriate framework? Yes, that's a lot of lines of code, it will probably be even more once put into C (almost certainly it will still be relying on someone else's libraries). But we might be talking about something that has more than one function. Unless the suggestion is that Matlab is just generally inadvisable for everything, but that seems unlikely.
(As an aside, I'd thought of mentioning boilerplate in interfacing different scientific libraries in C, but in checking a few things stumbled on the far more interesting snippet that Matlab licenses FFTW for its FFT implementation. It's a small world after all.)
Why is it the wrong tool for the job? What actually is the job in question? Are C programs bug free? Are they easier to verify?
I don't terribly like using Matlab if I'm honest. Quite a lot of its design makes otherwise trivial tasks pretty difficult if you want to turn something into an application. It's a pain to interface with. It wants to do its own thing when working with clusters. But it also sits on top of a heck of a lot of professionally designed routines for heavy maths.
I then spent several weeks converting them into C, finding numerous bugs in the process. For my trouble, I also got a 70-fold [sic] speed increase.
Neither of those are particularly arguments that Matlab isn't suitable for the application. It may well be sufficiently fast for what they're doing, and you would presumably have identified those bugs from code review or reimplementation between any pair of languages.
Yes, there's more to teaching than just giving the lectures. I know teaching fellows who are working flat out (evenings and weekends, and have been for some time) preparing online courses, demonstrations, student exercises and quizzes, answering questions ranging from "why doesn't this work?" (analysis) to "why doesn't this work?" (can't log in to moodle). Teaching materials also have to be updated, especially at the later year undergrad and masters levels as new methods and ideas come out.
Also, if you want to just take one set of lectures, repeat them ad-infinitum and fire everyone else, then you should be paying commission to the people who created them. There's usually quite a lot of "creators should get paid for their work" anger around here, but it doesn't extend to educators? Most decent universities in the UK have had this conversation with staff who have been working to put materials online that, no, we wont just sack you and use the stuff you've already created next year.
Well that's horrid. Regardless of the actual bug just looking at it a clear problem is relying on too-clever pointer trickery. Even without checking the standard my instinct would have been there aren't guarantees on layout for argv[] members, and I can't find any, making this implementation dependent even on occasions where it works as intended.
For people that just want free RHEL, and feel that Red Hat does not deserve money for their offering and the services that come with it, to help pay the 1000's of engineers that advance, they are now out of luck and can take their non-business to another place.
Oddly they've never paid me for a single bug report, even after I've escalated it upstream for them and run testing for patches.
Speaking of rocks... wonder what this means for Rocks? Current version is based on Centos 7.
We did Ubuntu for a while, it's always been a little odd in how it approaches some things (dash, no-root etc.) and I usually find RH more comfortable in terms of how things are configured generally. But these are mostly quirks when you compare to things like dropping Mate.
Yes, don't terribly like the look of this. Covid got in the way of our moving to RHEL8 and maybe it's just as well. We're currently covered by a site license, but we have to rely on what the community is willing to support, and many of the groups we work with target CentOS and Debian. If this causes them to drop CentOS then license or no it's increasing pain to stay with it. Coupled with the Gnome3 thing I'm wondering what IBM think the RedHat market is, or is this just the embrace, extend, extinguish cycle?
Fortunately you recall wrongly, actually asked and expanded on https://www.biblegateway.com/passage/?search=Luke%2010%3A25-37&version=NIV
36 “Which of these three do you think was a neighbor to the man who fell into the hands of robbers?”
37 The expert in the law replied, “The one who had mercy on him.”
Jesus told him, “Go and do likewise.”
Latin? Ο ιπποπόταμος, nominative plural οι ιπποπόταμοι ("oh, hippopotamus" sounds like a great clean curse, I may start using it). Although it's interesting to see my Oxford Greek/English dictionary in the English section still gives -muses or -mi, so I guess Greek speakers are not too worried about what we do to their words.
Spare a thought for those having to deal with neurology. The hippocampus (seahorse) is the small structure responsible for memory (shaped like a seahorse, because 16th century anatomists largely named structures based on what they'd had for lunch, see also "fornix" and be glad there isn't a part of your brain called "calamari"). What is the plural? If you tend to prefer following the Greek form you get hippocampoi, and everyone thinks you're weird, you can do the funny-foreign-word-into-pseudo-latin that we tend to use in English and get hippocampi, or you can take the bite-the-bullet-it's-an-English-word-now approach and arrive at hippocampuses, to which my only objection is it's just too long. The everyday get-out clause is to do what many do and shorten it to "hippo" and then you've got "hippos", but they wont let you get away with this in journal articles sadly.
The following is Plasma 5.15.5, YMMV...
VPN connections show up in the same list as hardware connections and just added with the same "Add a new connection" (the + button beneath the list of connections in the connections systems settings module),
GTK appearance settings: System settings, Application style (in the appearance category), GNOME/GTK Application style submenu.
Afraid I've no idea why you don't have a network settings icon, for me it's part of the system tray. (Can be disabled, but don't see why that would be a default, Edit panel, then configuration option on system tray, networks should be ticked.)
Pretty bad news considering Gnome3 doesn't work with X2Go. I see there's a Copr, not particularly impressed to see it dropped. We've stuck with RH for a while because of satellite and we technically have support (read: can access knowledgebase answers...), but remote desktop is now essential for us, as for many others.
GTK is LGPL licensed, meaning applications written against it do not need to be GPLed.
Qt is dual-licensed, there is a commercial license allowing you to develop closed source applications that use it or an option to use it under an open source license. My memory of this was that the only open source license option for Qt was GPL, meaning you could only distribute GPLed applications that used it (privately if not distributing you can do what you like), however it is currently available under LGPL, but LGPL 3.0 rather than the LGPL2 which still covers GTK. The difference between the two is not the same as that between GPL2 and GPL3, I'm not entirely sure what it is, to work it out you have to parse all the LGPL caveats about how you're allowed to distribute linked code. https://www.qt.io/download-open-source
Edit: I'm not 100% sure, but LGPL 3.0 may actually be more permissive rather than less. IANAL...
900,000m3, works out to be a ~120m diameter balloon, quite big. Which raises the question, maybe you could just (...) stuff an airbag beneath it instead?
(The cables are quite horizontal, which means they're carrying quite a few times the weight of the observatory, I guess somebody has thought about the tradeoff. Maybe it's jerk versus relatively static force that's part of the issue?)
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
