The APL Orchard

hyper-neutrino

00:49

idea: golfing language that's just an APL transpilation layer - using jelly chain structure, have two copies of each primitive function in monadic and dyadic form and that way you avoid the need for brackets or sub-dfns by making it more truly tacit :p

since arity depends on context / position within trains and thus you need brackets or use of composition to change the structure to get the arity you want (IIRC)

dzaima

if you're changing syntax of monadic vs dyadic behavior, it'd have to be more than transpilation

but yeah, i've thought about having an APL-like golflang with separated monadic & dyadic cases, i just have no clue if that is what Jelly already is (or what things there are to learn from it)

hyper-neutrino

well, Jelly chain structure is totally different from APL trains (from what I can tell) and the built-ins are more golf oriented ofc

but yeah by transpilation i mean like, you'd just parse the code and then just convert to equivalently working APL code

dzaima

i guess that could work, but scoping probably becomes a mess

(though i guess you don't really use variables in golfing)

hyper-neutrino

true. although jelly doesn't really have variables so you wouldn't need to deal with that i think

the inverse would be very annoying because of that but going from a jelly-like thing into APL should work without needing to worry too much about that, at least AFAICT

Bubbler

01:18

Some of the things in Jelly that might interfere with the transpilation are the register and very few (just one? or more?) atoms that modify the argument in place. We can simply avoid them in a new lang I guess

hyper-neutrino

yeah. the register is quite uncommon and is typically only used in weird situations to save a byte or two and can just be done without it

i think atoms that modify the argument themselves are just head/tail? not sure, but yeah, it really wouldn't be much of an issue for them to have no side effects, TBH

Adám

05:45

@RGS Not 2): Try it online!

@hyper-neutrino The tree-view is enabled by default on TryAPL: Try it!

hyper-neutrino

oh. nice :D

Adám

@hyper-neutrino ⎕←2-×⎕NC'⍺' gives 1 when the function is called monadically, and 2 when it is called dyadically. Try it online!

TryAPL

@Adám 3

Adám

Haha, very funny.

hyper-neutrino

.-.

Adám

05:53

(And 3 when ⍺ makes no sense.)

hyper-neutrino

@Adám oh, that's cool

Adám

See also APLcart

hyper-neutrino

oh, I see

so from randomly testing i see ⎕NC'x' seems to give 0 if undefined, 2 for a value, and 3 for a function

Adám

You do know we have ⋄]help ⎕NC, right?

TryAPL

@Adám help.dyalog.com/18.0/Content/Language/System Functions/nc.htm

hyper-neutrino

05:57

oh yeah

probably makes more sense than throwing random things at it to determine via experimentation

Adám

@hyper-neutrino The tc operator from the dfns workspace might be of interest too: Try it online!

hyper-neutrino

ooh, this is cool, thanks.

wait you can just do ⎕←--- without quotes?

TryAPL

@hyper-neutrino
┌─┴─┐
⊢ ┌─┼─┐
  - - -

hyper-neutrino

... oh wait that's what that means

Bubbler

How come there's ⊢ at the end

hyper-neutrino

06:09

⎕←(---)

TryAPL

@hyper-neutrino
┌─┴─┐
⊢ ┌─┼─┐
  - - -

hyper-neutrino

¯\_(ツ)_/¯

Adám

@Bubbler The ⎕← gets translated to ⊢ to coerce shy values (usually assignments) since ⎕← is otherwise prohibited.

hyper-neutrino

oh

⋄---

TryAPL

@hyper-neutrino
┌─┼─┐
- - -

hyper-neutrino

06:15

okay

Adám

However, not I think about it, it probably isn't really necessary.

Bubbler

⎕←----

TryAPL

@Bubbler
┌─┼───┐
⊢ - ┌─┼─┐
    - - -

Bubbler

And this is totally wrong

hyper-neutrino

oh, huh..

why did i stop stripping these agian? was it to make ⎕←a←3 work?

TryAPL

06:16

@hyper-neutrino 3

Adám

@hyper-neutrino Yes. The problem is on my end. I'll have a look at it.

hyper-neutrino

ah, alright :)

Adám

@RGS Guards, error guards, and literals.

Adám

06:54

@Bubbler Fixed: ⎕←----

TryAPL

@Adám
┌─┼───┐
⊢ - ┌─┼─┐
    - - -

Adám

No, I mean, will be fixed in a min.

⎕←----

TryAPL

@Adám
┌─┴─┐
- ┌─┼─┐
  - - -

Adám

@Bubbler @hyper-neutrino Fixed ^

hyper-neutrino

ooh, nice :D

Adám

06:58

You can even see my change on GitHub.

hyper-neutrino

ah. unfortunately i do not really understand :p hopefully eventually

Adám

@hyper-neutrino Quite simple, really: Down on line 103/104, 1 means "respect shyness", while 0 means "ignore shyness". Now, up at line 26, I remember if the value is shy (i.e. it does not begin with ⎕←) before I strip the ⎕← away.

hyper-neutrino

I don't think I really understand shyness - i'm looking at aplwiki.com/wiki/Defined_function_(traditional)#Shyness

RikedyP

@hyper-neutrino I disagree - while the help will definitely give you the answer, throwing random things is perfectly valid way of learning and I think you're less likely to forget if you've "discovered" something for yourself than if you just read it (not that remembering the nameclass numbers is particularly important)

@hyper-neutrino Most functions will print their result in the session even if you didn't ask for it (that is, a ⎕← or "print" function is not required) unless the function returns a shy result (or no result)

hyper-neutrino

07:14

@RikedyP well, the thing is when i did ⎕←⎕NC'⍴' i got ¯1 and I thought that meant "primitive" but looking at the page that's not actually what it means

TryAPL

@hyper-neutrino ¯1

hyper-neutrino

so i guess a bit of both is needed - experimentation to play around and looking at the actual docs to make sure i'm not misunderstanding :p

RikedyP

@hyper-neutrino That sounds like the way to go :P

Adám

Agreed.

@hyper-neutrino More experimentation to confirm could also be needed. E.g. if you tried $ and empty string and space etc. you'd probably conclude that it means "invalid".

RGS

@Adám yeah that doesn't leave much room to work with :P I can only get it to work from ¯1 to ¯1+2*15 or from -2*15 to 1...

Adám

07:17

Want my solution?

hyper-neutrino

@RikedyP ah, i see - thanks. so how would i define a shy func for example? like say i wanted {+/÷≢⍵} to be shy

RGS

Not yet, I'm going for a walk and I'll think about it while I'm outside. Walking gets my creativity to flow.

hyper-neutrino

tried doing {a←+/÷≢⍵} and it does what i want but is probably wrong

RGS

@hyper-neutrino nah nah, that's it.

hyper-neutrino

@Adám that's also true - i didn't really put enough things into it to get a meaningful result yet

RGS

07:19

@hyper-neutrino about shy dfns: mastering.dyalog.com/User-Defined-Functions.html#shy-result

Razetime

if you want letters variable names untouched you can also assign to _

hyper-neutrino

oh so that is just how you do it? lol okay

Adám

aplcart.info?q=shy dfn

hyper-neutrino

huh. okay, thanks. was worried it was a hacky workaround mostly, lol

Adám

CMC: make a dfn shy without affecting any variable names.

RGS

07:21

@RGS seems I'm missing some words on shy tradfns

Adám

(Off-hand, I can think of two major approaches.)

hyper-neutrino

i don't suppose {''⊣⍵} is a real answer? :p

oh wait no it still outputs a blank line lol

Adám

0 0⍴… is also not an answer.

Razetime

{{}result} ?

hyper-neutrino

oh i think i remember seeing that somewhere

Adám

07:22

@Razetime No, that's "no result".

hyper-neutrino

in the game of life video, i think?

Razetime

ah well

RikedyP

Is it the same but the inner dfn is shy using assignment?

Adám

"Shy" means that it doesn't display, but if you try to use the value, it is still there.

RikedyP

⎕←{{_←⍵}⍵}3

TryAPL

07:23

@RikedyP 3

Adám

@RikedyP Ah, that's a 3rd approach. Very nice!

Razetime

dzaima's APL always returns the last line unless ← is prefixed

RikedyP

I feel betrayed

⎕←'Good bot'

hyper-neutrino

lol

ironic

Razetime

⎕←'Good bot'

TryAPL

07:24

@Razetime Good bot

Adám

@hyper-neutrino Could the bot detect edits?

hyper-neutrino

⎕←'⎕←''yes i made sure to avoid this'''

TryAPL

@hyper-neutrino ⎕←'yes i made sure to avoid this'

hyper-neutrino

@Adám i have never worked with that in the past but i could try

Adám

It'd be kind of cool if it'd edit its own response when you edit your message.

hyper-neutrino

07:26

@RikedyP you can't reply with multiline code blocks, and every line in a multiline code block must be code for the whole thing to become a block

Adám

It'd also need to scan for edits to other messages that it hadn't reacted to before.

hyper-neutrino

back to the sandbox i go :D

Adám

07:52

@hyper-neutrino OK, fixed.

hyper-neutrino

ooh, very nice thanks :D

RikedyP

10:07

@kimmolinna Seen your tweet. Also thanks for the ClickHouse instructions which were forwarded to me. Are you using APL for processing and then a combination R stuff and Vega for presentation?

Paul Mansour

11:16

@RGS Oh yeah, mine too will only work in a specific range, but slightly different: negative anything, to positive 2*15 less 1.

Adám

11:41

@PaulMansour Mine too.

Paul Mansour

@Adám Do you count the diamonds as bytes?

RGS

@PaulMansour right, but that's a decent solution. I'm essentially saying I could only implement half of it, and the other half returns the opposite result

Adám

12:01

@PaulMansour Yes, but not f←, just from and including { until and including }

Paul Mansour

@Adám I can't figure out how you get 5 fewer bytes!

Adám

@PaulMansour Let me know when you want me to reveal mine.

Paul Mansour

@Adám Anytime

Adám

@PaulMansour Try it online!

Paul Mansour

@Adám Aha, I had almost the same, but I used two error guards where you used one, and used an 11 rather than a 0 for the first error guard.

Never occured to me to combine the two consecutive guards.

Adám

12:15

@PaulMansour Nice golf; my 11 can be 0 for only 19 bytes. Try it online!

Paul Mansour

@Adám Nice!

Adám

12:27

CMP: Is this bookmarklet useful?

javascript:((()=>{with(new XMLHttpRequest){open(`POST`,`tryapl.org/Exec`);setRequestHeader(`Content-Type`,`application/json;charset=utf-8`);send(JSON.stringify([0,0,0,prompt()]));onload=(_=>alert(eval(responseText)[3].join`\n`+`\n`))}})())

If you create a bookmark named "⍎" then, on any website, you can click ⍎ in your bookmarks bar pop up an input field. Enter an APL expression, and you'll get an alert with the result.

rak1507

to me personally, it wouldn't be, I always have a session open anyway, but in general it seems cool

kimmolinna

It's good that those instructions has been useful. I had almost one year layoffs so I had time to test different kind of things.
I'm using data.table in R as a database. I have done an interface for R in Dyalog. It's using Rserve binary server for data transfer between r and Dyalog. In Dyalog I use conga library for connection. The DyalogAPL is a swiss knife for me. I use it for formating a proper CSV-files, creating vl/JSON-files for vega-lite etc.
JupyterLab is a tool for presentations for me. I can easily show scripts and graphs in a same place. It's useful in CLA negotations because the…

Andy Aquino

12:47

@Adám oh, nice. Haven't seen that. Thanks! I'll bookmark this page.

@Razetime I downloaded Mastering Dyalog APL and read the index. It's a thick book and I just wanted something I could read like on a weekend. So that's why I started reading Learning APL with APLX. Although, I think Mastering Dyalog APL is really nice to get a deep dive into the language. I definitely will read it.

lyxal

13:18

Except for joining on newlines, I've written my first APL fizzbuzz!

(⊃⍪⌿){(⍵,,/(0=3 5|⍵)/'Fizz' 'Buzz')[1+⌈/(0=3 5|⍵)]}¨⍳100

It's horribly long and probably overly messy, but I'm very proud of it ;)

Razetime

@AndyAquino nice to know!

lyxal

@lyxal Razetime got it joined on newlines

↑⍕¨(⊃⍪⌿){(⍵,,/(0=3 5|⍵)/'Fizz' 'Buzz')[1+⌈/(0=3 5|⍵)]}¨⍳100

very cool

Razetime

protip: when using /(reduce) on an array, you usually need a ⊃ before it since a lot of the time it returns a boxed result

      ⍪{⊃(⍵,,/(0=3 5|⍵)/'Fizz' 'Buzz')[1+⌈/(0=3 5|⍵)]}¨⍳100
┌────────┐
│1       │
├────────┤
│2       │
├────────┤
│Fizz    │
├────────┤
│4       │
├────────┤
│Buzz    │
├────────┤
│Fizz    │
├────────┤
│7       │
├────────┤
│8       │
├────────┤
│Fizz    │
├────────┤
│Buzz    │
├────────┤
│11      │
├────────┤
│Fizz    │
├────────┤
│13      │
├────────┤
│14      │
├────────┤
│FizzBuzz│
├────────┤
│16      │
├────────┤
│17      │
├────────┤
│Fizz    │
├────────┤
│19      │
├────────┤
│Buzz    │

Adám

protip: when joining strings into a simple string, just use monadic ∊

@lyxal Congratulations, and welcome to the cult.

lyxal

@Adám are there any joining benefits? :p

Adám

13:29

Other than wielding the awesome power of array manipulation? Well, one more thing to drive away sleep…

Razetime

      ⍪{⊃(⍵,∊x/'Fizz' 'Buzz')[1+⌈/x←0=3 5|⍵]}¨⍳100

some improvements

Adám

@Razetime Since you have ⊃ there anyway, why not just use that to pick with?

Paul Mansour

@Adám This used to be significantly slower.

Adám

@PaulMansour What did?

Paul Mansour

@Adám super ravel vs disclose of cat reduce.

Razetime

13:33

⍪{(⍵,,/x/'Fizz' 'Buzz')⊃⍨1+⌈/x←0=3 5|⍵}¨⍳100

Adám

@PaulMansour I think ∊ special-cased, just like ,/

Paul Mansour

Could be now. But, regardless of speed, I think ∊ can cover up a lot of coding errors, that can bite you somewhere else. A lot of people use it because they have no idea why some of the items in the array are simple and some are not.

Obviously it does have its uses though.

Adám

@PaulMansour Agreed. What should have been instead is apl.wiki/Raze.

Paul Mansour

@Adám Ah, interesting.

@Adám It appears ∊ is now faster than ,/. I vaguely remember Roger discussing this a few years ago. He must have done something about it!

Adám

:-)

Marshall

13:47

@PaulMansour Probably my stackless traversal changes, or some later tweaks.

RGS

14:04

@Adám ahhhhhhhhhhhhhhhh clever

Paul Mansour

@Marshall Ah cool.

RGS

17:34

Can anyone help me with this? I created a link to a folder that didn't exist, and it didn't complain

(I assumed the folder would be created for me)

Then I created a function, the function was fixed, but it wasn't saved to the filesystem :P

oh nvm

I can't type paths

xpqz

17:52

I had a go at implementing the Knuth-Morris-Pratt substring search - or at least the prefix calculation (one of the Rosalind problems). I can't see an array way. Is there one? brilliant.org/wiki/knuth-morris-pratt-algorithm

Here's my attempt as a tradfn: gist.github.com/xpqz/1916eecdd39699f3411a87fb67fefc04

and as a dfn: gist.github.com/xpqz/e4467da656ae37024be917cf045ee01d

However:

      cmpx 'prefix d' 'pfx d'
┌→───────────────────────────────────────────────────────────────────┐
↓  prefix d → 2.1E¯1 |    0% ⎕⎕⎕⎕⎕⎕⎕⎕⎕                               │
│  pfx d    → 9.2E¯1 | +341% ⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕│
└────────────────────────────────────────────────────────────────────┘

The tradfn destroys the dfn.

This makes me sad.

dzaima

18:06

@xpqz does this work? avoids the 1↓

xpqz

      cmpx 'prefix d' 'pfx d'
┌→─────────────────────────────────────────────────────────────────┐
↓  prefix d → 2.3E¯1 |  0% ⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕│
│  pfx d    → 2.3E¯1 | -1% ⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕│
└──────────────────────────────────────────────────────────────────┘

!!!

dzaima

yay

xpqz

@dzaima So dropping the head of the list is the expensive op.

dzaima

@xpqz yeah, O(n)

(but of course with a very low constant factor because APL)

xpqz

Dropping the last object avoids the re-alloc?

Maybe?

dzaima

18:11

if the argument has refcount 1, it should, but i don't know if dfns are smart enough to recognize the last use of ⍵

@dzaima (fwiw, in CBQN (and J too I believe?) 1↓ is an O(1) operation, as slices of arrays are a core part of the array model)

xpqz

Note to self: APL isn't Scheme.

dzaima

@xpqz who'd've known

xpqz

:D

Anyway -- thanks; I'll try to remember that take-head-recurse-tail requires a bit of care.

dzaima

(trying to directly port to BQN is going awfully, with all the conditionals. maybe i should just look at what it's actually doing and go by that)

xpqz

@dzaima Here's a c++ version that's pretty clear: gist.github.com/xpqz/461d6a1564ab4e9be69374e665fa1031

(not one of mine...)

dzaima

18:23

no tail call recursion probably means it won't work much anyways :/

RGS

KMP is such a lovely algorithm

First time I learned about it I was 🤯

xpqz

It really is a masterpiece.

dzaima

what argument were you testing with? i think i got a working BQN solution

xpqz

Here's a short one for correctness:

      pfx ⎕←s
CAGCATGGTATCACAGCAGAG
0 0 0 1 2 0 0 0 0 0 0 1 2 1 2 3 4 5 3 0 0

The one I used for timing is very long.

dzaima

@xpqz yep, gives ⟨ 0 0 0 1 2 0 0 0 0 0 0 1 2 0 0 0 1 2 3 0 0 ⟩

oh wait, that's wrong

xpqz

18:35

@dzaima The big one is here: github.com/xpqz/rosalind/blob/main/data/rosalind_kmp-1.txt

dzaima

ok fixed, gives the correct thing now

xpqz

You need to drop the first line, and join the others

dzaima

seems to take about 120ms

nathan rogers

xqc?

dzaima

@dzaima (code)

@dzaima (Dyalog takes 200ms, so not too much faster than it :/)

dzaima

19:10

@dzaima and i can't seem to improve that at all..

(perf report - 8.6% of time is just spent freeing scopes, and 9% is block calling overhead)
https://dzaima.github.io/paste/#0VZFBT8MwDIXv/RW9cI0Sp2kbTrABEggYGnCOQppp07q1dBu/H9tZhrjle3bi95xW1PqqLA9hGKNbTTEWlWgqVFan/ax3QSHrFjn@@P7p8aPQogZEPJZGX5ez@Y2qLaqGmnYdpEtaaIMchn3nAhDSm@MmbBNC/TeCWJKJ1TBxGYTV6bG7ZAlEay51hWgIN@7hVc2RKuqO3@kqVOcSUElR6at3O9UVSliTnUt2bppakqzOcmVTIDAtZN0fpiO9rFLutZ@mZEoJo3nU7XLpRkJuPw5jauet0Wig0brOthZIYCmfo4SZ@qNPy0GWl9UpRGX56uKNmxXl812XWjkfJ4JkXVMgScOCcpOfYiGFbf6jzj45BwqS93fyPULLtiPuOtJ4FJLz9/uPl03G6MN6R1Dlf0I3iPxtjv91CFtSJF9@Xsw/kRqb3VbsFsD8Ag#BQN

code_report

19:45

Not sure this has been posted here yet: github.com/interregna/arraylanguage-companies

If you know any companies using APL (SimCorp, etc) ... would be awesome if ppl could open Pull Requests or even just Issues asking for the companies to be added

I would love to see a comprehensive list

rak1507

aplwiki.com/wiki/Talk:Jobs already exists which could be useful

also aplwiki.com/wiki/List_of_famous_APL_users but there's not much on it

Transcript for