CMC: Given a strictly positive integer as left argument and a vector of 2-element (integer string) vectors as right argument, return ⍳⍺ but with elements divisible by ⊃¨⍵ replaced by 2⊃¨⍵, respectively, while letting later elements of ⍵ overrule earlier ones.
20 f (3 'Fizz')(5 'Buzz')(15 'FizzBuzz')
┌─┬─┬────┬─┬────┬────┬─┬─┬────┬────┬──┬────┬──┬──┬────────┬──┬──┬────┬──┬────┐
│1│2│Fizz│4│Buzz│Fizz│7│8│Fizz│Buzz│11│Fizz│13│14│FizzBuzz│16│17│Fizz│19│Buzz│
└─┴─┴────┴─┴────┴────┴─┴─┴────┴────┴──┴────┴──┴──┴────────┴──┴──┴────┴──┴────┘
18 f (2 'Fizz')(4 'Buzz')(6 'FizzBuzz')
┌─┬────┬─┬────┬─┬────────┬─┬────┬─┬────┬──┬────────┬──┬────┬──┬────┬──┬────────┐
│1│Fizz│3│Buzz│5│FizzBuzz│7│Buzz│9│Fizz│11│FizzBuzz│13│Fizz│15│Buzz│17│FizzBuzz│
└─┴────┴─┴────┴─┴────────┴─┴────┴─┴────┴──┴────────┴──┴────┴──┴────┴──┴────────┘
@Adám Is this output fine? https://tio.run/##nZDPSsNAEMbvfYrxtC0k1OoTiCAKnqwvEJrtH2gSSVdUOl5UShuboojiUXrKoeBFj17yKPMicbIxlRAE0zlkJ1/mm/y@tc6Gpn1lDb2eKS@VdG1pJ8mYJo9Gk4IXCmfcjgUKI45aNP@qOzR9ZSmOKPxsGBTcCFM0KYwcmjzQfNqk2VJ/u46jeiubNXb0@dTgNXWavVEwoWDZyMaSbmoMFxTcUrii8CN@32UPLZ7bJ/v8PD08aic9nhnzCx9C0P1dF1JrrQcCYQ/60rKlD9y6nupztxaOrZGCjjc8d1wWfQmAIvWYPwW/7R@C1rTnYODzMi6EbSgUguPx7o7n8v2p7EeZpy1ZtPWI710UPapvKSFGMFBba1FsEKl6oOpxNgiTRcmTYDEHllLo@8oBsciLpQR6Ok@ARX4s0evpnB6L7Fgmx2rolcArcf8TW9S@AQ "APL (Dyalog Extended) – Try It Online"
Ok, I hope this is correct: https://tio.run/##nVDLSsNAFN37FeNqEkiorV8ggii4sv5AaKYPaBJJxxcdN1VKGztFEcWl1E0WQjd26WY@5f5IvJmYSgiC6V3k3pw5584545z1bffa6Qcdm11x5rvMTZIhjB@tGkQvIKc4DqmglorrMPsy1DJtHkxe8UDFOINcmRZEI2rTGsjYg/EDzCY1mC7wWK5uVGzUM7rV0P3JNFCWakcXPOx5KjZBvjdUjNcYMH2DaAzRwszUSTvdJ@cQ3YL8APmplru4CubPzZN9/J4eHjWTDnKG@IONUri/a5NUutUhVJA90mWOy0KCox/wLk5r4NgZcNIK@ueej2DICBE01dg/RX7HPwCNac1BL8RlWILskEIJ4gW4uxX4@L48uyjTNBmCrqaEwWVRw7sOp3RAenx7DdINIlUPVD3OBmGyKHkSUcwhSin0e@UGRdGvKCXQ7DyBKPoXJfeanbsXRe@i7FxUs17JeCXf/7RNt74B
@Adám ⍳0 to get ⎕io was an idiom when I was a lad.
@ngn How would that work with real-time applications? APL was used for quite a few of those. At one time, British Airways decided where to put the curtain between economy and club using an APL program that read real-time data from BABS (the BA real-time reservation system).
@RomillyCocking i liked this! the idea of 'software as stored [executable] knowledge' resonated with me, though why it's connected to APL or array transformations more than other languages / data processing was less clear (if that was your point)
The REPL is a standard html textarea which means individual characters cannot be formatted.
@RomillyCocking Why would one use that instead of ⎕IO? Sure, it saves you 1 byte, if on its own, otherwise you may well need (⍳⍬) which is longer, and all this at the cost of being very obscure to the reader.
@Adám just remember that you may get called out to fix it at 3AM when a critical batch job is interrupted because of bad input from an external source, and a few $100M worth of trades required to balance the pension funds of tens of thousands of people aren't happening, as required by law. At that point, clarity and the ability to fix it in a few seconds while half-asleep is quite valuable.
@MortenKromberg I work with a treasury system used by a very large proportion of banks, and I know very well the "middle-of-the-night-this-end-of-day-job-must-finish-before-start-of-day" problems.
@MortenKromberg Banks are all about managing risk. So they tend to be very wary of changes. That means we have to be as well. It gets tricky when the software is 60+ million lines of code.
That's a long-winded way of saying: Yes, golfed code is frowned upon.
The bulk of Dyalog's revenue comes from organisations who deliver asset management systems or operate asset management services.
Other significant users are the people who design & manage oil and lubricant production using APL, and the worlds largest electronic patient journal system which manages records for more than half the Swedish population (fun fact: I computed that about 2% of the population of population of Stockholm have user ids in this system as they are somehow involved in Health Care). The latter system is registered as a "medical device".
Much as we are language geeks at Dyalog and enjoy working on proposals to extend the language, we need to be a little bit careful about making changes to the interpreter, since all the carefully crafted application code rides on top of everything that we do.
@MortenKromberg We (Cocking and Drury) sold mainframe APL into British Airways. If it went down, the whole airline moved from daily profit to loss that day.
In my case, APL is not used in our product. It's mostly C++, Java. And a much of GPU code in CUDA. However, I like using APL when doing things like working on performance statistics logs, etc.
@RomillyCocking That's a bit of an exaggeration I hope? It's not like you'd deploy a new version without testing?
I mean, we had to stay on Java 8 for way too long, until all our dependencies had been updated to handle all the backwards-incompatible changes in Java 11. It was a pain, but we did it after a couple of years :-)
Of course we and they tested new versions. The point Morten and I are making is that APL has been and still is used for business critical applications. In that environment customers, and responsible vendors, are rightly very cautious about change.
Fun story: when BA were about to switch to a non-IBM but allegedly compatible mainframe they paid for the US author of the component file system to fly over on Concorde to discuss the risk of using the new hardware.
@RomillyCocking Theys upport all the libraries shipped with a iven version of RHEL, until that version is EOL. I believe Python 2 will be supported until 2025 or something like that.
We might get rid of the python code altogether before then. It's not that much.
We use some libraries RH ships (openssl, python, some others). But we have a large number of third-party libraries shipped as part of the product. But in those cases we provide support for them, and they are only used by us (i.e. the libraries are not exposed to the customers)
The Python stuff, however, is for the most part customer code. Python can be used by the customer to extend the trade evaluation models used (i.e. computing different payoffs).
And as you know, customers don't like it when things change so we can't just switch to python 3 for payoff calculations.
Leaky Relu is ⍵ if ⍵ is positive, but 0.01 × ⍵ otherwise. I can write it, no problem, but my solutions are not as concise as I would like. The solution needs to work for ⍵ of arbitrary rank, of course.
Another golf-ish problem from MENACE code: I am currently using this to index the last dimension of an array of unknown rank, but I am sure there is a better solution.
@RomillyCocking That's fair. Btw, have you seen my proposal for a Select (a.k.a. "sane indexing) function, usually denoted ⊇? With it, this would be just ⊇⍤1
@EliasMårtenson Indeed, and the experimental array languages are of great interest and value to us. Apologies in advance to y'all for moving slowly and only occasionally st... er, adopting any of your ideas.
@RomillyCocking I find it interesting that when Replicate has a Boolean left argument, we call it Compress. When Select has a left argument which is a permutation vector, I'd call it Permute.
@Adám I link to those in the latest (unpublished) version of the MENACE book. I think I will now publish (though ti;'s still not quite complete), so y'all can give me more of your excellent feedback.
@RomillyCocking without downtime? i have no experience with such migrations but i would expect you'd still need realistic data for testing the new system, and backups made just before flipping the switch. so being able to export the data is still very important.
@RomillyCocking i've never read it
@RomillyCocking i wish i'd read this before i answered :)
@ngn The site has now been updated with a clarification. Jonas Stensiö did achieve the same score as you, but he sent in all of his submissions at the end of the year, and so wasn't eligible to the other monthly listings. Your score was 270 because you were strict about the scalar vs vector, which is why your score was adjusted to 269. We agree that ⎕SD is wrong and will (hopefully not forget to) update the text file and 267→268 soon.
@Adám for this particular case probably no, because you corrected it now. of course, the most honest thing to do would have been to publish all participants' submissions.
That BA system used BABS (the reservation system) but was an entirely separate piece of code. It sucked in real-time data from BABS and emitted decisions but stored no history.
Same for the fuel-buying system, which read real-time jet fuel prices around the world and worked out whether it was cheaper to buy fuel where you were (and pay the fuel cost of flying it back) or to buy when you reached your destination.
Those two apps between them had a major effect on BA's profitability day-to-day.
IIRC, during an aviation crisis where all other airlines lost money, BA was the only one that didn't have a loss, as they were able to quickly adapt their APL code to the new reality.
@Adám Correct. And then they tried (several times) to rewrite the relevant systems in Java. I imagine that they have by now, which imay be why they don't do so well these days :)
@RomillyCocking Right, with enough effort, any APL code can be rewritten in Java, but a side effect is basically setting the code in stone ― it won't be flexible going forward.
@Adám No, that's not true. Since the efficiency of a team drops quadratically with the number of developers, and requirements are constant changing, there are some apps that cannot be implemented in less productive technologies.
Sadly the managers who need to understand that never do.
Such projects usually fail because, as the huge effort required becomes clear, the managers who never understood 'the Mythical Man Month' do the equivalent of asking nine mothers-to-be to make a baby in a month.
I need help D: I thought APL source code (for dfns, namespaces, ...) would preserve its original whitespace if saved in text files..?
I opened a fresh Dyalog interpreter session and linked the root workspace # to a folder, typed )ed f and wrote a basic function with some whitespace in there.
ESC fixed it, and it shows in my file explorer as f.aplf, but )ed f again shows that the whitespace is missing.
Was ON, I tried turning it off and did everything again for a new function. Similar issue. The text file now has one less space (ASCII 10) per line of code.
@Adám speaking of the Dyalog comp (am doing the beta testing) -- guidance states that one should seek "proper array solutions, not loops", which is the APL way. But what does that actually mean for the comp? Avoid ∇¨ and possibly \/ and ⌺?
@xpqz I'd like @Brian to answer that, as it might also need clarification in the instructions. Feel free to include that question in your beta feedback.
@xpqz I'm not involved in judging the solutions, but as a general rule the best solutions are Pure data-parallel operations, because they will most often consume less memory AND CPU. SHARP APL & J style operators (rank, stencil) try to preserve simple arrays and are often optimised by interpreters, so they should be "cool".
@xpqz APL2 style operators (like each) often lead to less efficient solutions. Of course, there is a trade-off between the efficiency of reading&writing the code, vs the efficiency of execution - nested arrays may lead do "pointer chasing" by the interpreter, but often allow much more readable solutions.
@xpqz Some of the APL2 style operators are also optimisable, at least for primitive operands. At the other end of the spectrum are actual loops, which can still be justified. Judging will always be quite subjective.
@MortenKromberg yes, that was the reason for asking, really -- obviously "readable" is in the eye of the beholder, but as a relative newcomer to Dyalog and APL, the intuition of what is efficient and what is a performance death-trap does not come easy.
@xpqz When I write code that I think might be performance sensitive, I use the "cmpx" function from dfns a lot - or the ]runtime user command to verify my intuition. And complain to the developers when my intuition fails ;-)
<raoul> Hi, I'm used to J but starting something in Dyalog APL. If I want to make a lib, is it correct to just stuff my code into a named workspace, and then import it? Or is there some other idiomatic way? Thx!
raoul: Welcome. I would look into storing things in text files. If your things are currently in the workspace, do ]link.create # /path/libname -source=ns
That should create a directory libname which has text files in it, and can be imported with ]import libname /path/libname
And such a directory of text files is well suited for SCM like Git.
@Razetime The main reason is that you can't highlight a textarea without a bunch of hacks. That's also why dzaima's Paste page has separate edit and view modes. The syntax tree view ("Explain") does highlighting.
@Razetime I should probably do more to improve it. It wouldn't be too hard to give the primitives short explanations on hover, and link to documentation pages on click.
@RomillyCocking raoul: You should only need .NET to auto-sync file system changes, e.g. after a git pull ― the message is just a warning.
If you're developing something by yourself, from inside APL, that's not a necessary functionality, even if you use SCM. And you can always ]link.refresh to re-sync if needed.
raoul: Yes, you should be able to ]add ⎕IO though this may not work if you're using an older version of Link. You can simply create a ⎕IO.apl file in the directory that has the content 0
@xpqz What we look for is appropriate application of array-oriented thinking. For instance you could implement summation using recursion or reduction; the judging committee would view the recursive solution as "loopy" whereas reduction is array-oriented. I mention "appropriate application" because not all problems have clean, array-oriented solutions. I hope this helps.
@Brian I sort of get it -- I guess it's a little bit confusing if this is a 'problem solving' competition, or 'beauty pagent'. For example -- without any spoiling -- certain problems have trivial regex solutions, but I suspect that's not what the judging panel would look for or expect. Other problems have solutions that can be solved almost completely using functions found in 'dfns' -- again, what say the judges about that?
@RomillyCocking very neat project will bookmark and read when I have time/interest. 2 years ago i tried some ANN stuff in gnu apl. I got a basic example working, but did run into some difficulties.
1.) because dfuns use tradfuns behind the scenes in gnu apl, i noticed some 'spooky' behavior where ⍺ ⍵ seemed to change values outside of their scope. could not reliably recreate 2.) some numerical calculations seemed to give slightly off values. this was likely due to me not understanding how gnu apl stores values. notably qr decomposition i had issues with (it was an experimental feature when i started). 3.) it was not fast (hard to optimize when the userbase/contributor base is so small and the project is so mature)
i would like to port my math fns to bqn after i get the erlang interpreter working. i looked at some bqn code yesterday and realized that i have not internalized the language like i did w/ apl (yet)
@xpqz Regex solutions are allowed. (Personally, I think anyone who thinks APL is hard to read has never spent much time with regex :)). As for the dfns workspace, it's been the bane of my existence in developing problems for the competition - find problems that John Scholes hadn't already solved in dfns. Participants are allowed to use code from the dfns workspace, but when we judge we take into consideration the originality of the solution.
btw not trying to put gnu apl down i have a lot of respect for Jurgen and gnu apl but its hard to both be an apl2 alternative and also incorporate ideas from the last 40 years
@Adám Are folks aware that neither registering nor newsletter subscription forms work? (The message says they've been "notified that sending e-mail failed", so hopefully they weren't notified by e-mail :P )
<raoul> Following Adám's advice above, I've tried to force ⎕IO←0 on import of my lib by creating an ⎕IO.apl file with content 0 in the Linked dir, but the system still tells me ⎕IO=1 even after Link.Refresh. I must be missing something...
<raoul> To clarify, I linked a lib as ]Link.Create lib, where the lib dir contains a file ⎕IO.apl that only contains 0
Finally, I got custom operators working in Kap, using the syntax proposed by @dzaima
The main difference compared to standard APL is that the function arguments are passed as first-class functions, and are not bound as regular functions in its implementation.
<raoul> How can I branch without eval'ing the false branch? For example, I'd like to test whether ⍵ is numeric and then if it's positive. {(0<⍵)∘⊣⍣((1=2|⎕DR)⍵)⊢'b'}'c' yields a domain error.
raoul: This sounds worrying. Can you try using the latest Link from the master branch at https://github.com/Dyalog/link with installation instructions at https://github.com/Dyalog/link/blob/master/help/Installation.md#standard-installation
Those kinds of rules are completely subjective and have nothing to do with Python as such. They're also frequently used to batter other Python programmers' code with. Same with the whole 'I'm more Pythonic than you!'.
Don't get me wrong -- I write Python for a living, and it's good at what it does.
@xpqz I guess I understand your point of view. I take the Zen of Python as a generic set of guidelines to get you in the right mindset for coding in Python. I agree it is arguable whether that's of any value or not.
@xpqz True, but it is in a pseudo-Zen style. It is intended to make you think. E.g. "Complex is better than complicated." → "Am I over-complicating this?"
E.g. one of my colleagues has a tendency to want to implement an "All-singing all-dancing API" rather than specify a strict but simple API.
moon-child, why would you refer to numpy as a language? You, now, and some other people in here sometimes seem to talk about numpy as if it were a language in its own right.
<moon-child> @RGS the term 'programming language' is somewhat loosely defined. I would also call unix a programming language, because it allows a certain way of structuring programs and systems in a way that wouldn't be possible otherwise (in other types of operating systems). I think numpy plays a similar role in that it allows a certain way of structuring programs that owuldn't otherwise be possible in python
<raoul> @Adám Ok, the behaviour is as expected with latest Link: ⎕IO is 0, while lib.⎕IO is 1 :-)
<raoul> @Adám So, I guess I am supposed to import my lib names directly into the current workspace for lib.⎕IO to overwrite #.⎕IO and it should all work, right?
<raoul> @Adám Sorry, I meant to write above: ⎕IO is 1, while lib.⎕IO is 0 ! Still, seems to work OK.
<raoul> @Adám The stock Dyalog version distributed on the Dyalog website has Link do the wrong thing, though: both ⎕IO & lib.⎕IO are 0
<raoul> @Adám both ⎕IO & lib.⎕IO are 1 ! I mean 1 !
<raoul> @Adám damn! I'm too used to indexes starting at 0...
raoul: Importing a library can't (and shouldn't normally) affect the root ⎕IO. However, your import or link your main project to # rather than to a sub-namespace: ]link.create # /path/myproj and myproj's ⎕IO will take effect in the root.
raoul: If you never want to use ⎕IO←1 for your own things, consider setting the environment variable DEFAULT_IO=0.
the first two I probably think are good because they have simple, readable syntax ... because they're "easy to learn" ... so they have 1-based indexing
<moon-child> lua was originally intended as a configuration file format. And then it slowly evolved pl features. I think if the creators were making it from scratch they probably would have done 0-based indexing
<moon-child> apl at least you can change the io, and most non-apl apls are 0-based
