The APL Orchard

@-moz-document unobservable-document() {
  @media (prefers-color-scheme: dark) {
    :root {
      /* in-content-page-{color, background} for dark theme. */
      background: #2a2a2e;
      color: rgb(249, 249, 250);
    }
  }
}

nathan rogers

 1 2 5\!10
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 1 1 1 1 1
0 1 2 3 4 0 1 2 3 4

Adám

@nathanrogers That's ⊤

nathan rogers

this looks like split, at least its called split in the docs

Adám

@nathanrogers Yes, that splits into digits of a given (mixed-)radix.

ngn

@nathanrogers split is "separator"\"string"

Adám

7:15 PM

I really dislike that difference between data types there.

ngn

i really like the symmetry and analogy with other uses of / and \

nathan rogers

1 3^"abcdefg"
bc
defg

then ^ is cut

Adám

@ngn How does one split a numeric list on 1 0?

nathan rogers

I think you would just do a &

(&1 0)^list

 (&1 1 0 0 1)^"abcdefgijk"
a
bcd
efgijk

Adám

      a
1 0 0 0 1 0 1 1 0 1 0 0 0 0 0 1 1 0 1 1
      1 0⍷a
1 0 0 0 1 0 0 1 0 1 0 0 0 0 0 0 1 0 0 0
      1 0(⍷⊂⊢)a
┌───────┬─────┬───┬─────────────┬───────┐
│1 0 0 0│1 0 1│1 0│1 0 0 0 0 0 1│1 0 1 1│
└───────┴─────┴───┴─────────────┴───────┘

@nathanrogers I don't get it. How do you do ^ ?

nathan rogers

7:20 PM

I'm not sure I follow. indices ^ list

& is ⍸ so (&bitArray) ^ list

Adám

@nathanrogers Given my a above, how do you get the final array above?

ngn

@Adám why not "given my 1 0⍷a above"? you're asking 2 questions in 1: what to pass to ^ and how to compute 1 0⍷

Adám

@ngn Because my original task was based on the missing functionality to "split on numeric sequence" when the parallel with the existing "split on character sequence" is obvious.

rak1507

cmc: decrement the first number in a vector

Adám

-∘1@1

rak1507

7:26 PM

there's a 4

ngn

@Adám i see

@rak1507 in apl?

rak1507

yes

Adám

I was looking for an unusual scan, but I can't seem to find one.

rak1507

not a scan

Adám

@rak1507 That's short enough to brute-force.

rak1507

7:34 PM

here's a hint: I didn't just come up with it randomly

ngn

@Adám but not important enough to bother actually doing it :)

Adám

@ngn Wait, are you content with living without solving this‽ Who are you, and what did you do to ngn?

rak1507

the underlying idea could save ~a byte or so in some situations, maybe

ngn

@rak1507 ?@0 <- there! it worked

rak1507

haha

ngn

7:39 PM

@Adám well, there are easier methods to learn the answer

trying lots of random things, actual logical reasoning, beating rak1507 with a rubber hose

rak1507

I like the sound of the first 2 more than the 3rd one...

Adám

@ngn You could also offer to pay him. You know, capitalism… Everything has a price.

rak1507

10000 pounds per golfing tip :D

ngn

@Adám rubber hoses are cheaper than developers :)

Adám

I know where he lives.

rak1507

7:43 PM

oh no

nathan rogers

yeah I have no idea what I'm doing. @ngn how would you do that?

you're right its really 2 questions

ngn

@nathanrogers as you said, a boolean mask can be converted to a list of indices with & (where)

nathan rogers

right, but how would you do something like a windowed comparison?

Adám

@ngn That's the easy part.

nathan rogers

well, that's the question you asked, how do I partition

ngn

7:45 PM

for 0 1⍷ maybe eachprior for this particular case, like >':

it may need some adjustment with 1_ or appending something

Adám

@ngn OK, but in the general case.

ngn

@nathanrogers "sliding window" is n'

nathan rogers

(&,/0 1>':a)^a
10001
01
101
000001
101
1

close, but no stoagie

ngn

@nathanrogers no no, just >': without a left arg. it means "greater than each prior", i.e. "is the current item a 1 and the prior a 0" for bools

rak1507

here's my ⍷ implementation {*/a,'{x#0}'(#y)-#'a:(!#x)_'x=\:y}

idk what I'm doing

ngn

7:48 PM

@Adám in the general case something like 0 1~/:2'a

nathan rogers

that doesn't work in shakti

ngn

(sliding window of size 2 over a, and then does 0 1 match each element on the right)

Adám

@ngn You'd want the length of the subsequence as argument, no?

rak1507

oh, sliding window is cleverer

nathan rogers

I was trying to do the sliding window but it isn't working in shakti

ngn

7:48 PM

@nathanrogers yeah, i don't remember what sliding window was there

nathan rogers

:C

ngn

maybe n': ?

nathan rogers

what's the word for it in k?

that's each prior

 2':a
2':a
^
!rank

ngn

@nathanrogers dyadicfunc': is eachprior. int': might be something else.

rak1507

{x~/:(#x)':y}

ngn

7:50 PM

@nathanrogers ok, it's not :(

nathan rogers

sd

I was close

ngn

@ktye do you know how to do sliding window in k9?

rak1507

cmc: take elements from a vector until reaching the first duplicate item

ktye

@ngn mavg:{(y-(-#y)#(-x)_y:+\y)%x}

Adám

@rak1507 2 7 1 8 2 8 1 8 2 8 becomes 2 7 1 8?

rak1507

7:55 PM

yes

just found a 9 which is shorter than the cool 10 I had

Adám

Bad test case, though.

ngn

@ktye that takes advantage of the associative properties of +

rak1507

1 2 3 4 1 9 becomes 1 2 3 4

Adám

@rak1507 I have 8 so far.

rak1507

ooh

Adám

7:56 PM

@rak1507 Simple vector?

yep

7

:O

@ktye we needed something equivalent to n,/ in apl, e.g. f[3;"abcdef"] -> ("abc";"bcd";"cde";"def")

Marshall

6 in BQN.

That translates to 8 in Dyalog 18 with ⊢⍤/.

Adám

7:59 PM

@Marshall Same idea. That'd be ∧\⍤≠⍛/⍨ in extended, but you get ⍨ for free in BQN.

@Marshall Yes, but then there's the ∊⊆ golfing trick for 7.

ktye

@ngn what is the original question?

rak1507

I have a 4 I think

ngn

@ktye split a sequence of numbers on the occurences of a given sequence of numbers

nathan rogers

each previous also doesn't seem to be what I want

Adám

@ktye My complaint about "abc"\ splitting a character list on "abc" but 1 2 3\ not doing the same for a number list.

ngn

8:01 PM

@ktye like "sep"\"string" but for numbers

nathan rogers

,':a
(10;01;00;00;10;01;10;11;01;10;01;00;00;00;00;10;11;01;10;11)

a
10001011010000011011

Adám

@rak1507 ∊∧\⍤≠⊆⊢ in case that wasn't obvious by now.

@rak1507 On your CMC?

nathan rogers

second index is 01? that doesn't look like the right output @ngn

rak1507

yes

@Adám it was, and it helped me

ngn

@nathanrogers the current item is x, the prior item is y

Adám

8:03 PM

@rak1507 Ooh, I have an idea.

rak1507

I would never have thought of it without you mentioning ⊆

my 'cool 10' was {⍵/⍨≠\≠≠⍵}, using 3 ≠s :D

Marshall

I see it too. Very lucky you can't have a length-0 result.

rak1507

≠≠ is interesting I wonder if it would ever be useful

ngn

@nathanrogers so, the pairs returned by ,': are reversed with respect to the original order

ktye

@Adám just use a proper k: ktye.github.io/…

Adám

8:05 PM

@ktye proper=7?

ktye

@Adám no, ktye.

Adám

@rak1507 ⊃≠⊆⊢

rak1507

yep

so cool

nathan rogers

(&(1_{&/0 1=x}',':a),0)^a
(1000;101;10;1000001;1011)

48 mins ago, by Adám

      a
1 0 0 0 1 0 1 1 0 1 0 0 0 0 0 1 1 0 1 1
      1 0⍷a
1 0 0 0 1 0 0 1 0 1 0 0 0 0 0 0 1 0 0 0
      1 0(⍷⊂⊢)a
┌───────┬─────┬───┬─────────────┬───────┐
│1 0 0 0│1 0 1│1 0│1 0 0 0 0 0 1│1 0 1 1│
└───────┴─────┴───┴─────────────┴───────┘

ktye

there was an answer for just that from geocar on hn some time ago.

rak1507

8:06 PM

did you figure out decrementing the first number yet

ngn

@nathanrogers {&/0 1=x} could be {0 1~x} or even (0 1~)

Adám

@rak1507 No. @Marshall Any ideas on decrementing the first element of a vector in 4 chars?

rak1507

he should read up if he wants a clue

nathan rogers

@ngn I'm not sure why I need the 1_...,0

kind of annoying

ngn

@nathanrogers think about it: each item from the list is paired up with its predecessor, and there's an implicit "seed" before the first item

Adám

8:09 PM

@rak1507 Clue to that CMC?

Marshall

@Adám No, but 1⊸↓⌾/ is a cool BQN solution for natural numbers.

rak1507

@Adám yep

nice, ⌾

nathan rogers

(&(1_(0 1~,)':a),0)^a
(1000;101;10;1000001;1011)
@ngn

Adám

@Marshall That is very cool.

ngn

@nathanrogers so, we are matching the pattern from its end. a 1 in the bitmask means the end of a match. that's why the 1_ .. ,0 - because we want a 1 to mean the beginning of a match

nathan rogers

8:11 PM

I really just wanted a 2 window

:P

and with this approach, it wouldn't work for arguments greater than 2

ngn

@nathanrogers tell that to shakti. other ks have it.

nathan rogers

there isn't even a rotate

or if there is there's no name for it

I could do an iterative take drop scheme to implement rotate

Adám

@nathanrogers Why not concatenate the take after the drop?

nathan rogers

or just use take drop to get windows

ngn

@nathanrogers trivial to implement, rarely needed

nathan rogers

8:14 PM

@Adám I don't know what you mean

Adám

@nathanrogers Implement the equivalent of APL's ↓,↑

      3(↓,↑)'abcdefgh'
defghabc
      3⌽'abcdefgh'
defghabc
      ¯3(↑,↓)'abcdefgh'
fghabcde
      ¯3⌽'abcdefgh'
fghabcde

nathan rogers

well that's what I'm saying, chunking seems like a pretty common need in array programming, a rotate could do that iteratively but it has neither windows nor rotate, so I guess the alternative is just to use take and drop like you showed that's waht I said here chat.stackexchange.com/transcript/message/57277651#57277651

idk why shakti doesn't have those things. clearly he doesn't use them or think other people use them all that much

ngn

@nathanrogers or ideally indexing. indexing is very fast.

@nathanrogers this is how my (ngn/k) sliding window is actually implemented: S A2(win,K("{y(!x)+/:!0|1-x-#y}",x,y))

ktye

ifrom memory: x=\:y (eq each right) returns a list, than shift each row by the row number, now a column of all ones, means everything matched. split on that.

i think the comment was about: it's not necessary (a primitive for that) because it's obvious for an array programmer.. i don't know if i agree.

rak1507

8:30 PM

my linalg isn't great: why is ⌹⌹3 2 ⍴ 4 5 6 erroring? (⌹3 2 ⍴ 4 5 6)+.×3 2 ⍴ 4 5 6 is the identity matrix, so 3 2 ⍴ 4 5 6 should be the inverse right? idk

maybe it's a floating point thing

it works if you do ⍉⌹⍉ (⌹under⍉)

ngn

@rak1507 matrix multiplication is not commutative, so there's a left inverse and a right inverse

nathan rogers

@ngn (&18#1 0)^|1_|1_,/2#'!10

2 window indices

rak1507

@ngn ah

nathan rogers

well not really

I've missed the last element

rak1507

@Bubbler chat.stackexchange.com/transcript/message/57277085#57277085

cmc: given a vector of all unique elements find where elements are equal to their index (0 1 2 5 4 6 8 -> 1 1 1 0 1 0 0 when ⎕IO←0)

ngn

8:49 PM

@rak1507 too easy, ⍳⍨=⊢

rak1507

I have a 3 :D

ngn

∊=⌸ ?

rak1507

yep

oh well that one didn't last very long

Adám

CMC: Given a list of length 2 or longer, answer the number of applications of monadic ≠ needed to reach 1 1 0 0 0 …

rak1507

15 atm lol

Adám

8:57 PM

That's some pressure.

ngn

if only apl had converge-iterate..

rak1507

≠1 1 0 0 0... is not 1 1 0 0 0 so you'd need something along the lines of dfns.traj

this is a very interesting challenge

I think I have an 8 but I'm not sure that it works

Adám

9:14 PM

@ngn Even if given a function for ≠, could you do this easily in K?

rak1507

it works on some random inputs, so I think it works

how long is your solution?

Adám

I didn't write one. Just posted the CMC to keep you busy.

rak1507

oh

ngn

@Adám it doesn't converge, so probably not as easily as using \ directly

rak1507

oh my solution doesn't work anyway

Adám

9:14 PM

@ngn My point.

rak1507

it finds a number of ≠s that work but not necessarily the minimal number

Adám

@rak1507 24 so far.

ngn

i have 17

this bug with ≠¨ is annoying

Adám

@ngn You can pretend that's fixed: Z←⊢⍤≠

ngn

17: {⍵≡2>⍳≢⍵:0⋄1+∇≠⍵}

Adám

9:26 PM

@ngn That's very nice and clean.

ngn

..and long :|

rak1507

I don't think any method will get shorter than that, I think {(a←2≠/2↑⍵)∧2≡≢∪⍵:1⋄a:3⋄2} works which can be golfed but not short enough

ngn

@rak1507 didn't you say you had 8?

rak1507

yes, it was wrong

nathan rogers

0 {⍵ ≡ 1 1↑⍨≢⍵: ⍺ ⋄ (⍺+1)∇≠⍵}⍳9

ngn

9:29 PM

ah

nathan rogers

is this not it?

oh you can cumulatively sum instead of using ⍺ I guess

Adám

@nathanrogers Yes, that's pretty much exactly what I had.

nathan rogers

sorry I wasn't looking at chat :C

Adám

@nathanrogers I didn't post mine. It was

0∘{⍵≡1 1↑⍨≢⍵:⍺⋄(1+⍺)∇≠⍵}

nathan rogers

:P

rak1507

9:30 PM

my 8 was 2+∘⊃2≠/⊢ which finds a number of ≠s that work but it fails on something like 1 2 1 2 where it should be 1 not 3

ngn

@nathanrogers you like inserting spaces everywhere even when golfing :)

nathan rogers

t⍤t yes I do love my whitespace

ngn

i win :P

rak1507

now figure out how to decrement the first item in a vector

nathan rogers

@rak1507 {⍵+1↑⍨≢⍵}⍳9
2 2 3 4 5 6 7 8 9
?

rak1507

9:33 PM

decrement, not increment, and in 4 chars

ngn

@rak1507 are you sure it's just apl and not any dzaima, "extended", etc stuff?

rak1507

yes

nathan rogers

{⍵+¯1↑⍨≢ ⍵}⍳9
0 2 3 4 5 6 7 8 9

rak1507

that's one way to do it...

Adám

Hm, about 20 million possibilities.

Any dyadic operator would seem too expensive, byte-wise.

nathan rogers

9:36 PM

I don't have all the golfing idioms down yet :C

I don't know where to take this one

rak1507

this isn't a golfing idiom which is why everyone's stuck ;)

nathan rogers

4 characters ?

i'll bet it uses encode

rak1507

no

nathan rogers

or some such nonsense

rak1507

it uses something that was discussed earlier

nathan rogers

9:36 PM

sign probably

rak1507

that should be enough of a hint to get it

are you going to try narrowing it down one primitive at a time

Adám

Eureka!

rak1507

did you get it

Adám

@rak1507 That's what I did, and it didn't take long to find it.

coltim

is it something like +~<? (plus not grade-up)

or minus =S

Adám

9:38 PM

@rak1507 Yes, but your hint from earlier didn't lead me at all. Only after I found it, did I realise how it was a hint :-)

@coltim What? That doesn't look right.

So obvious, now that I think about it. D'oh.

ngn

@coltim apl's "not" tolerates only 0 and 1

rak1507

haha yep wonder if the secret bit has any golfing applications, seems like it might be useful in niche scenarios

Adám

Fun fact: I had eliminated all but 64 primitives by the time I found it.

coltim

@ngn it's like speaking only-somewhat mutually intelligible dialects, heh. I think a k answer could be {x-~<&#x}

Adám

@coltim Yeah, that is indeed very similar to my solution.

nathan rogers

9:42 PM

I did it

ngn

@coltim also: @[;0;-1+]

nathan rogers

@rak1507 ⌽{1=≢⍵:¯1+⍨⍵ ⋄ (⊃⍵),∇ 1↓⍵}⌽⍳9

rak1507

lol

ngn

my best is 5: -⌿↑⎕1 or -∘1@0

nathan rogers

⎕1?

Adám

9:44 PM

@nathanrogers ⎕ 1

ngn

@nathanrogers it's stranding the evaluated input (⎕) and 1

rak1507

@ngn what was being discussed when I posted the cmc

Adám

@rak1507 What should it do on an empty vector?

nathan rogers

no, I don't get it

Adám

@nathanrogers a←⎕ ⋄ a 1

rak1507

9:46 PM

@Adám ⍬≡f ⍬ is fine, I don't know what else it would do

ngn

@rak1507 many things. splitting on numeric lists? k9's ^? ..

coltim

@ngn or {x-~<x?x} maybe?

rak1507

pick one of the many things, if it works, great, if it doesn't, try another one of the many things

Adám

@rak1507 Right, and that works for my solution. But ngn's two solutions fail.

nathan rogers

oh, that's odd

Adám

9:46 PM

@nathanrogers What is?

nathan rogers

how that expression works

Adám

@nathanrogers How so?

ngn

@rak1507 ⍷⍨ ?

rak1507

bingo

nathan rogers

I get it, but why wouldn't you just write it some other kind of way

{-⌿↑⍵1}⍳9
0 2 3 4 5 6 7 8 9

ngn

9:47 PM

ah, right ⊢-⍷⍨

rak1507

⊢-⍷⍨ but yes

Adám

@nathanrogers Because using ⎕ instead of ⍵ saves 2 bytes from the {}

nathan rogers

not if you count input

Adám

@nathanrogers What?

nathan rogers

you have to input a valid argument

Adám

9:49 PM

So?

ngn

@rak1507 clever cmc

rak1507

thanks

Adám

Indeed. That was a true puzzle.

nathan rogers

so that's part of the input

Adám

I don't get what you're getting at.

nathan rogers

9:50 PM

-⌿↑⎕1 this depends on what the user types in, it isn't a standalone solution is all I'm saying

rak1507

generally in code golf either a function or full program is acceptable, so using ⎕ for input is

nathan rogers

but it depends on the users input

Adám

@nathanrogers {…} also depends on the argument that the user gives.

nathan rogers

I can't just run -⌿↑⎕1

a user has to interact with it

rak1507

you can't just run ⊢-⍷⍨ either

nathan rogers

9:52 PM

no but it is a function

rak1507

you have to give it an input, either by ⎕ or by passing it as an argument

nathan rogers

nah, one is dependent upon mutation

one isn't

Adám

@nathanrogers You can look at -⌿↑⎕1 as the body of a tradfn too. The header is irrelevant and not used or counted.

Alternatively, you could pipe -⌿↑⎕1 into the interpreter and it act as an interactive program.

rak1507

it's just convention that a function or full program are acceptable

ngn

if we could do a full-text search for 1↑⍨≢ and suggest ⍷⍨ - imagine the number of thankyous!

rak1507

9:53 PM

personally I do prefer a function solution though

nathan rogers

Its cool and everything, I just don't think that it is a useful lesson to be learned from golfing

rak1507

@ngn doubt it's that common

Adám

@ngn It isn't the same for higher-rank arrays.

ngn

@rak1507 i used it only a couple of days ago - for nathan's number spiral, i think

rak1507

well, you're welcome :)

Adám

9:57 PM

CMC: Given argument, but with all elements replaced by 1s. restricted-source: No literals.

rak1507

=⍨?

Adám

Fails on nested arrays.

≡¨⍨

Yes. Too easy.

1⊢

9:58 PM

just a slight modification of the well known =⍨

Adám

@nathanrogers Fails on non-scalars.

nathan rogers

{1}

Adám

Now, do it in the most interesting fashion, while sticking to 3 bytes! popularity-contest

@nathanrogers Fails on non-scalars.

nathan rogers

      {1}((1 2)(3(4 5(6 7))))
1

seems to work

Adám

> 1s

rak1507

9:59 PM

{1} 1 2 3 4 5 doesn't

Transcript for