The APL Orchard

Adám

01:00

0

A: Create a binary matrix based on information given by an "Excel" table in the APL language

First, let's get your questions out of the way: how could it interact with matrix t, after all the rules need to be researched in matrix t? Right, this might be involved. I think it is easier to construct the result directly, rather than to modify a pre-made matrix. how to insert the values...

Razetime

04:27

I'm trying to make an ASCII house:

Try it online!

I need to add a star at the top point

whats the most golfy way to do that

Improved a bit

Razetime

04:47

What I got so far

brgal

06:11

@Adám The simplest expressions I learned first (+/ vec ; ×/ vec ; ⌈/ vec ; +\ vec ; ×\ vec), I still find them beautiful. Maybe not "impressive" but... I like Roger Hui's answer to a "related" question back in 2012 groups.google.com/g/comp.lang.apl/c/BbyYv7kgvEw/m/rTnUKDMwH9cJ

@Adám One that I found beautiFUN is ⍎⌽⍕⌈*○≡⍬ Maybe just because I like 42.

Adám

09:58

@user14835830 Kimmo, I'll get you access to chat here soon.

RikedyP

@Adám No but the concept is pretty stunning I think

@phantomics Wish I'd saved this - seems to be gone now

xpqz

11:50

Given a nested matrix of integers, I'd like to get to a vector showing which cols have elements containing only 1s. The best I've managed is:

⋄ ⊢a←2 2⍴(1 1)(0 1)(1 1)(0 1) ⋄ ∧⌿(∧/¨)⍤1⊢a

Dyalog APL

xpqz

Is there a better way?

dzaima

@xpqz are the parentheses and ⍤1 not pointless?

xpqz

Yes, they are :/

Razetime

Trying to use breadth first search here:

Adám

12:01

@xpqz Are the matrices Boolean?

Razetime

Try it online!

why is this giving an INDEX error?

xpqz

@Adám yes

Adám

⋄ ∧⌿~0∊¨2 2⍴(1 1)(0 1)(1 1)(0 1)

Dyalog APL

@Adám 1 0

xpqz

nice

why didn't I think of that?

Razetime

13:13

Hello again

Writing an answer to this:

11

Q: Will it float?

The challenge Given a 2d string representing the bottom of a boat as an input, you must determine whether or not the boat will float. This 2D string can be in whatever format is most convenient. (String with newlines, list of string, list of list of chars, etc.) Print a truthy value if it will f...

Try it online!

is this properly golfed?

also, is this algorithm correct?

RikedyP

14:19

For today only (until the next one): Pythagorean triple day! tinyurl.com/yam9ep9v

xpqz

AoC day16: gist.github.com/xpqz/e44e172089386b3c6bd00216ac59e19b

Razetime

15:37

@xpqz github.com/razetime/AOC2020-APL/blob/main/day16.apl my day 16

ngn

@Razetime you're addicted to ≠⊆⊢ :)

Razetime

Anything to not use regex

@ngn It might actually qualify as my favourite APL expression

ngn

cmc: extract the numbers from a string, ignoring all non-digits. e.g. 'departure location: 49-258 or 268-960' -> 49 258 268 960

xpqz

@ngn regex allowed?

ngn

@xpqz sure

xpqz

15:57

'(\d+)'⎕S'&'⊢

Wezl

working on a non - regex solution

xpqz

Or for actual numbers, ⍎⍕'(\d+)'⎕S'&'⊢

Wezl

I need something like reduce but that preserves structure

so 1 1 1 0 0 1 0 smt 1 2 3 4 5 6 7 gives 1 2 3 0 0 6 0

ngn

@Wezl ×

Wezl

oh, I was just thinking and but not only for booleans. thanks

Razetime

16:03

@ngn ⋄'departure location: 49-258 or 268-960'(⍎¨∊⊆⊣)⎕D

Dyalog APL

@Razetime VALUE ERROR

Razetime

⋄'departure location: 49-258 or 268-960'(∊⊆⊣)⎕D

Dyalog APL

@Razetime 49 258 268 960

Razetime

execute isn't allowed i guess

Wezl

@ngn ok {⍎⍵×⍨(⍵∊'0123456789')} would work if × worked for characters

xpqz

16:06

@Razetime nice!

I'm stealing that.

Razetime

I think it's somewhere in the tips thread

phantomics

@RikedyP @RikedyP, you can still get the image here: i.postimg.cc/pVj8rM01/APLLogo1.png

ngn

16:22

@Wezl ×⍨ is ×. '0123456789' is ⎕d. i think one of the meanings of \ can be helpful for what you're trying to do: {⍎∊(⍵∊⎕d)\¨⍵}

Wezl

what's monadic epsilon?

ngn

@xpqz no need for ( ) in the regex. btw, the usual golfing rules allow only a "full program" or a function, so it should be either ⍎⍕'(\d+)'⎕S'&'⊢⎕ or some kind of train or dfn or derived or function, etc

@Wezl flatten, e.g. (0 1)2(⊂⊂2 2⍴3 4 5 6) -> 0 1 2 3 4 5 6

Wezl

I guess partition is what I needed, like in Razetime's excellent answer

ngn

@xpqz an idea for shortening that: instead of searching for runs of digits, replace the nondigits (\D) with spaces and ⍎

xpqz

Like so: ⍎'\D+'⎕R' '⊢⎕ ?

ngn

16:35

@Razetime so, that's either ⎕(⍎¨∊⊆⊣)⎕D or (⍎¨∊⊆⊣)∘⎕D. to get rid of the parens and make the whole thing a train: ⍎¨∊∘⎕d⊆⊢

@xpqz yes, without +

RikedyP

@phantomics Yeah I can thanks - also later after that message I was able to see the original ones so not sure what was up with that

@phantomics Anyway I'm personally leaning towards this as my favourite one, if only because people can adapt the boxed-with-arrow motif for all kinds of purposes, with any colours and fonts they like - the only downside then is what would be the canonical "official" APL logo? Not sure I'd go with the pixel-style for that, personally, even though I think your designs look great

@Razetime I am also very fond of this, even if sometimes I have to reach for (~∊⍨)⊆⊢ for multiple delimiters

voidhawk

16:51

Finally got my day 16 solution to something reasonable: here

Razetime

@ngn nice

@RikedyP interesting

ngn

cmc: given a number ⍺ and 4 sorted distinct numbers ⍵, find out if (⍵[0]≤⍺)∧(⍺<⍵[1])∧(⍵[2]≤⍺)∧(⍺<⍵[3]), i.e. ⍺ is in one of the closed-open intervals ⍵[0 1] and ⍵[2 3] (flexible on open/closed, ⎕io, and swapping ⍺⍵)

phantomics

@RikedyP Thanks, I also had adaptability in mind when designing the logo. The pixel styles can be adapted with many different color palettes. I could also create a vector version of the boxed arrow design that wouldn't have the pixelated look.

ngn

17:27

my solution for day16

voidhawk

17:39

@ngn Neat. I somehow always forget that you can just use multiplication for selection with boolean arrays :(

Better version of my solution with some of @ngn's tricks

Razetime

@ngn if they are 4 distinct sorted numbers, isnt the answer always 0?

the number can only ever be in one interval

Marshall

@Razetime Should be ((⍵[0]≤⍺)∧(⍺<⍵[1]))∨((⍵[2]≤⍺)∧(⍺<⍵[3])) I think.

ngn

@Razetime oops.. you and marshall are right

Marshall

I have three characters unless I confused some of the flexibilities.

Also they don't have to be distinct.

ngn

@Marshall me too :)

@Marshall duplicates in ⍵ could cause problems if someone chooses to use closed-closed intervals

the two intervals could overlap

Marshall

17:59

Oh wait, there are at least two different three-character strategies. My first works for arrays ⍺ and my second doesn't.

ngn

@voidhawk interesting part2. i thought it should always include some loop equivalent to ⍣≡

phantomics

18:17

Is there any try-it-online service implementing NARS2000?

Marshall

@phantomics Answer from last time seems to be no.

Adám

Dec 10 at 20:54, by dzaima

@Adám it runs fine under wine

voidhawk

@ngn Yeah, in the general case that would be true, but the inputs for this problem are constructed such that there's never any "branching" when we pick the next field assignment

dzaima

18:39

@Adám that's performance/not-erroring-wise. Actually using it is horrible.

Adám

Can you elaborate?

dzaima

@Adám it ignores the linux keyboard, forcing its own left-alt one; the various hover popups randomly remain after mouse has left, or even after tabbing away. those 2 things (plus in general the windows 2000-esque UI) have me mostly not want to use it for long enough to find more problems

Adám

@dzaima OK, so that's basically the same as on Windows. It also forces its own keyboarding on me there, including combos that I cannot access, like Alt+number.

dzaima

@Adám oh yay

~~who will sacrifice themselves and write a NARS2000 RIDE interface~~

Adám

The interpreter doesn't speak RIDE protocol, though. I'd be so nice if all APL interpreters spoke the same front-end language, so you could just mix and match front- and back-ends… Won't happen, of course.

dzaima

18:50

@Adám my message was about making it able to speak the RIDE protocol

Adám

Oh. Well, NARS2000 is open source, so sure: Any volunteers?

dzaima

@Adám exactly.

phantomics

Thanks, I'll try it under Wine then

ngn

@Adám plain text on stdin and stdout is the best universal language

dzaima

@ngn cannot support autocompletion, multiple windows, etc. of course you don't need those, but you're not the only user of all APL impls

phantomics

18:54

Another thing, is the ⎕ct comparison tolerance part of the APL standard? Is it considered an essential feature?

dzaima

@dzaima (but i'll agree that it'd be good if every impl at least had a simple in→out REPL)

Adám

@phantomics It is part of the standard. ngn will tell you it is not needed.

dzaima

(well, ngn will also tell you floats aren't needed, so)

Adám

Latest iteration with slight adjustments of line widths and positioning:

ngn

@Adám correct :) it breaks transitivity

@dzaima correct :) but sadly they are part of the metal

dzaima

18:56

@phantomics but i'll also say that comparison tolerance isn't really needed. maaybe a modifier for checking equality, taking an operand of the acceptable error, but one can survive without that too

Adám

@dzaima I once wrote a blog post draft for such an operator, but it was never published. Interested?

dzaima

@Adám ooh, interesting, yeah. thought it'd be too verbose for an APL audience, but i guess (maybe) not?

phantomics

If it's part of the standard I'll probably implement it, currently I just wrap CL's =, /= and other functions with the APL ones

Since April has rationals, there's a problem with ⍣ fixpoints for functions that can have rational output; usually they never converge and loop infinitely

⎕ct will help with that but it feels less elegant

dzaima

@phantomics making it apply to rationals seems very weird to me

phantomics

That's why I want to see how NARS2000 does it

To use it I'll have to convert rationals to floats and compare them using the ⎕ct

Adám

19:09

@dzaima A Deviant Operator.

dzaima

@phantomics what it does

@Adám being a fan of ⎕CT what ←0 oh :D

phantomics

I could just set a special condition in the ⍣ implementation to handle ⍣=

dzaima

@phantomics at that point an approximate equality operator would be much simpler

phantomics

I like the deviant operator concept

@dzaima That's from NARS2000?

dzaima

@phantomics that's from Adám's article

phantomics

19:15

I meant is your paste from NARS2000?

dzaima

@phantomics yeah

phantomics

Got it, so ⎕ct doesn't affect rational comparison

dzaima

also note that in NARS 1000000000000000000=1000000000000000001 is false - it has a separate 64-bit int type (adding a .0 to either constant makes it return 1)

phantomics

Anything that big entered in April generates a bignum, CL numbers can get really huge

ngn

@dzaima the right approach! two numeric towers

dzaima

20:07

@dzaima (oops used BQN lingo there - "modifier" is "operator")

Adám

@dzaima Yeah, I noticed.

dzaima

@Adám that's an interesting alternative to what i had in mind - essentially just 1.5 (.2⁼) 1.6 for .2≥|1.5-1.6 and hoping everything else isn't needed because it shouldn't

Adám

@dzaima I can definitely see things like ⍳ and ∊ being useful.

dzaima

@Adám heh, my first thoughts were that ⌊⌈ and maybe ≡ would be useful but ⍳ and ∊ definitely being completely useless

@dzaima i guess they'd be useful in deduplicating similar numbers, wherever that's useful (heh, monadic ≠±n)

rak1507

20:42

github.com/rak1507/Advent-Of-Code-2020-APL/blob/main/… I have to say I'm quite pleased with my solution for advent of code from today

phantomics

20:59

One other question: APL and Lisp have slightly different formats for nested arrays. I added an option to April that converts nested arrays to Lisp-format when they are output. This option is currently called "unformat-output." What it does is convert from APL-style nested arrays where every nested array is wrapped in a 0-rank array to BQN-style output where n-rank arrays can be contained directly inside another array.

Can anyone think of a better name than "unformat-output"? It's not very descriptive.

dzaima

@phantomics why do you need to wrap nested arrays in rank-0 arrays? The only difference between APL and BQN style arrays is the existence of enclosed scalars

phantomics

In APL every element in an array must be scalar, hence nested arrays must be wrapped in a 0-rank scalar array

Adám

@dzaima Enclosed simple scalars, and non-array atoms.

dzaima

@Adám 1) right; 2) that's purely a terminology difference as far as i'm concerned

@phantomics not sure why that'd be needed. Arrays can just contain other arrays

phantomics

That's my understanding, otherwise x[1 2;1 2]←2 2⍴9 doesn't work correctly

Marshall

21:05

@phantomics I call what you're referring to a 0-cell rather than an element (which is the thing a 0-cell contains; APLWiki has my definition of element because I wrote that page).

(But APLers tend to use "element" and "item" inconsistently and don't even notice).

phantomics

Got it, each element being a 0-cell

dzaima

@phantomics regardless, those still shouldn't show up in the internal representation

phantomics

If those aren't part of the internal representation, a lot of complicated logic is needed so that functions will handle nested arrays correctly

Like ⊃ and ⌷

dzaima

@phantomics x[1 2;1 2] is equivalent to 2 2↑x, and there's no reason for (2 2↑x)←2 2⍴9 to not work

phantomics

The choice was either to create a structure that was orthogonal to APL's, which Lisp's array model fortunately makes straightforward, or to write logic into every function that would treat a nested array element as if it were scalar in the right situations

dzaima

21:10

@phantomics ⊃ should work fine ('ab' ≡ (⊂1 1)⊃2 2⍴⊂'ab'); ⌷ is a weird case, but again, it's much like ↑ in that it only rearranges elements

Marshall

@phantomics I think the problem is that you're trying to get 0-cells by taking array elements in Lisp. You shouldn't do that: get a 0-cell by selecting the element then enclosing it.

If the element is a simple scalar then enclose needs to detect that and not enclose it, because in nested APLs enclosing a simple scalar doesn't do anything.

dzaima

@dzaima (forgot that ⌷ can be used to get non-0-cells ಠ_ಠ; it still should absolutely not care about anything about what the elements of ⍵ are)

phantomics

@dzaima The point of that example is that the logic for assignment becomes confusing in some cases without the scalar element model

@Marshall That's what I did in the past, I can't recall off the top of my head but there were definitely situations where that model had problems

dzaima

@phantomics assignment logic shouldn't be that different from regular logic

phantomics

I'll look over those functions again, I already have an internal function analogous to ⊂ that encloses arrays but not scalars

Marshall

21:18

@phantomics Simple scalars, right? ⊂⊂1 2 should be different from ⊂1 2 even though it's a scalar.

phantomics

Yes, non-array scalars

I think my adoption of the 0-rank wrapping was driven by observations of the behavior of ⊃, ⌷ and [] indexing. Also the fact than when you employ it, the logic for assigning from one array to another becomes simpler

so v[1;1]←3 4⍴⍳9 produces an error instead of assigning the enclosed array

dzaima

@phantomics what enclosed array where?

phantomics

But the logic for assignment of nested arrays vs. elements from an array is also straightforward

@dzaima That's just it, the array to the right of ← isn't enclosed

But it's easy enough to just insist that an array to be assigned as a nested array must be enclosed

Marshall

@phantomics Typically that expression should see that v[1;1] would always have empty shape but 3 4⍴⍳9 has shape 3 4: they don't match, so it's an error.

phantomics

Yeah, that logic is simple

I think there were other factors driving the decision I made at the time, but I can't remember them.

After I get some of the current changes done I'll create a source branch without the 0-rank nesting and see if it can be made to work

If that's not a problem then it'll be a big improvement, no incongruity with regular CL nested arrays

Adám

21:51

@Marshall Didn't you once tell me that ∨/⍷ was optimised?

Marshall

@Adám I sure hope not. There's definitely an issue for it, maybe under RFEs. I barely touched ⍷.

Adám

Ah, ok. I probably began writing X(∨/⍷)Y instead of ∨/X⍷Y in anticipation of that.

Transcript for