The APL Orchard

@Marshall right, i did read that, but i think i want to leave too fancy things to after the core impl is actually somewhat finished (i guess left primarily is adding 8-bit/16-bit/1-bit arrays)

Adám

Meh, I don't know how to control the computer's priorities. Doesn't drop below 55 ms for ?1e4⍴1e5

Marshall

@dzaima Converting booleans to 1-byte indices will never have to change, and Compress is pretty trivial given that as a utility function. So I think you could do that now if you wanted to. Definitely important to avoid accumulating too much code that has to be changed when there are more types though.

dzaima

@dzaima i was thinking monadic ↑ would be quite fast, but it's actually still the runtime version. {m←↕0 ⋄ {s←𝕩⊑x ⋄ 1⊣m∾↩ s+ 0⌈´ m/˜ s>𝕩↑x}¨↕≠x←𝕩 ⋄ ⌈´m} gives 182ms

Marshall

It's using ↕⊸⊏ instead of dyadic ↑, because the provided ↑ doesn't have to handle a high-rank right argument.

dzaima

with that, 85% of time is spent in / (followed by isum of 6.14%, which slash uses to compte the result length)

with -march=native, it's 90.3% in /, 3.43% in isum, 3.8% in > (but doesn't end up noticeably faster)

but imma go sleep (pushed the native ⌈⌊-folds)

Razetime

6:27 AM

APL problem solving competition question

Adám

6:49 AM

@Razetime Well, clearly not copied from there, and requires much more checking.

Razetime

7:09 AM

a bit more I guess

but it's the same concept otherwise

LdBeth

7:31 AM

should I avoid the above link if I'm participating dyalogcompetition?

Adám

@LdBeth Not yet, at least. For now, there's only a Jelly solution. Probably won't help you much, as the problem in the competition is really trivial, and is mostly just about knowing the primitives and how to combine them. Solutions in other languages will hardly help…

LdBeth

Well, not yet throughly read that solution, but the link does remind me I can came up a better one than my current answer. ;)

Razetime

10:22 AM

RIDE idea: when one RIDE window is opened, other RIDE windows are also brought to top.

Adám

@Razetime Opened or gets focus?

rak1507

10:42 AM

@Adám hmm not sure if I'd necessarily say my solution was 'trivial', maybe I'm missing something

Razetime

11:25 AM

@Adám gets focus

Adám

11:44 AM

@Razetime Maybe suggest it on GitHub as an option. I'm not sure it is universally desirable.

Razetime

12:08 PM

@Adám after a bunch of clueless eight cllicking, there an optino in the right click menu in the title bar which says "Above session" which does the exact thing

Adám

@dzaima @rak1507 This looks ridiculously short (by ngn, of course): |/{x,|/(-1_y<*|y)*x+*|y}/,\

rak1507

neat

dzaima

12:24 PM

@Adám gives a different result to my solution to some random inputs

Adám

Huh. Using ngn/k?

dzaima

@Adám yep

that should give 12 for 5+7

Adám

Hm.

dzaima

prepending a 0 seems to fix it

@Adám I think the idea of that one is pretty much the same, just that a reduce is used instead of mutating a variable, and × is used instead of replicate (as the max-reduce afterwards doesn't care)

dzaima

12:53 PM

comparisons of my old fn, a new one like ngn's, and versions where they multiply instead of compress:
https://dzaima.github.io/paste/#0vZK7TsNAEEX7fMU0KLb8yiQhpIKKv4AKOTbFUtggF8ENQRaYGPEQoocmBVKKQIOUxvmT@RJm1xsETZQHwlp5tTN377ljGUA@vYiyu77gFxVvQDeXYNDwyhMmUP4K/Vg1PigfJKqJXBU2FyG2oMFKKQbhUTGCeNdgqYMmZfdJWo6omND1S1I5pDWoeOIveLPn5WiBmk7eYcfKPB/QcErFp7Jnqf3LmQt6lnKMlD2yMPUMyi84rWlzhy/bBz8JYnWCSr@kf8iOe@i3qXhHf1sXD6Ozk9Nj4YNzBPVeBGGdN6G2oDoF8qTVGzw1@YNACJQ9QMfF/XLchHPZaGwB3T6tsRaAhAa13Z05yGm2VgEtniSYT4JuiwHIAAuxu@4kmy8VSnyHQh0KrG7nv0Lpr/MF#APL

dzaima

1:08 PM

faster

@Adám (for h:10000?100000 it takes 238ms in ngn/k, so slower than both Dyalog and CBQN (CBQN's at 41-42ms now, so faster than Dyalog for whatever reason))

Marshall

Is CBQN using slices for ↑? Could be part of it.

dzaima

@Marshall i'm comparing to my best function in CBQN, which doesn't use monadic ↑

(but it is using slices for dyadic ↑ definitely)

Marshall

I meant the dyadic one.

dzaima

right. That probably explains why it can be faster than Dyalog

{-1_x}',\0,h alone is 150ms, but that still leaves 90ms on non-slice stuff

Marshall

Also, 1e4 elements probably is small enough that constant overhead still matters some, and ∾↩ could be better than ,← somehow. They're both amortized constant, but maybe it reallocates less or does it quicker.

dzaima

1:20 PM

oh right, and ngn's version probably doesn't do in-place appending at all

(i'm assuming ngn/k doesn't have an equivalent of LOCU)

xpqz

1:35 PM

@RGS Thanks for your proofing notes.

RGS

1:56 PM

@xpqz np. You can react to them in the issues and I'll keep working on the next chapters.

goof

@dzaima How slow is this? {v←s←⍬⋄⌈/s⊣{s,←⍵+⌈/0,(v>⍵)/s⋄v,←⍵}¨⌽⍵}

dzaima

@goof a bit faster than mine :)

(and faster than CBQN D:)

even faster

what - is ⌈/0, an idiom?

Adám

@dzaima Not that I know of.

Adám

4:33 PM

Announcement: APL Campfire in 90 minutes.

0xACE

5:00 PM

What am i doing wrong: {⌊⍵,⍵-⌊⍵} 7.62 i get 7 0, expected 7 0.62

rak1507

APL is right to left, so ⌊(⍵,⍵-⌊⍵)

0xACE

also "We can use a name in the same line in which it is defined. In production code it is best to avoid this unless an expression is very short." why are sameline gets not recommended? and why are the appropriate for short expressions rather than long?

rak1507

@0xACE if you do inline ← it's not as obvious where some variable declarations are if you're reading the code

0xACE

oh i see, thanks

rak1507

it's a matter of personal preference either way

0xACE

5:04 PM

Nah i understand. i thought it was a runtime efficiency limitation and it didnt make sense to me why shorter expression were faster. But if it's only a readability thing I totally understand the point. I'd advise a comment is put there to indicate "... because it is difficult to tell the value of a variable in long expressions"

user

Pretty nice basic rundown of k syntax here on Hacker News

0xACE

Damn i missed that apl campfire :(

rak1507

I knew what it was before I even clicked

geocar is great

0xACE

5:29 PM

{×/0≠(1+⍳⌊⍵÷2)|⍵}21 17 i expected this IsPrime to be a Monadic function but i get a RANK ERROR. What am i doing wrong?

dzaima

@0xACE you forgot an ¨

0xACE

Hmm, why is it required for my function, but not others?

would i have to ¨ all of my ⍵ in my functions?

rak1507

look at what ⍳⌊21 17÷2 gives

dzaima

@0xACE you don't need it for rank-agnostic functions (which are all your regular arithmetic functions), but here you're doing ⍳10 8, which isn't the same as ⍳¨10 8

0xACE

{{×/0≠(1+⍳⌊⍵÷2)|⍵}¨⍵} 21 17 would this be a bad idea?

or is there a better approach?

(not referring to the isprime function, jsut the idea of applying `1 like that)

dzaima

5:35 PM

@0xACE you could just {×/0≠(1+⍳⌊⍵÷2)|⍵}¨ 21 17

Adám

@0xACE No you didn't. About to start in 25 mins.

dzaima

{…}¨ works exatly the same as {{…}¨⍵}

0xACE

thing is i didnt understand "rank-agnostic functions" so I expected my function to work

@Adám oh, nice, I'll keep in in the background :)

dzaima

alternatively, you could do {×/¨0≠(1+⍳¨⌊⍵÷2)|⍵} 21 17

0xACE

That's the part i dont get, why does i and / need 1 ?

I tried your examples and noticed `i10 8 and `i`1 10 8 gave different results

dzaima

5:39 PM

@0xACE because they don't act the same way on arrays and scalars

0xACE

how are scalars and arrays differentiated in APL?

Adám

@0xACE They are not. Scalars are zero-dimensional arrays. See apl.wiki/scalar

dzaima

@0xACE right, and that's just how ⍳ is defined to be. On a single number argument it gives 1…n, and on a vector argument it creates an array containing all indices in that array

@Adám well, that's already a difference

Adám

?

dzaima

(also, every time i said scalar, i probably meant simple scalar)

Adám

5:44 PM

@0xACE apl.wiki/Scalar_function may be useful reading.

0xACE

I'll save both of those links for later, because I've overextended my timeslot for APL for today... deadlines, deadlines, deadlines...

dzaima

@0xACE If you called a function with an argument of [1,2] in JS/Python/whatever, you won't necessarily necessarily get the same answer as calling it with 1 and 2 separately. Same in APL.

(but the difference is that, in APL, a given function is much more likely for a function to implicitly map than in other langs)

0xACE

so in this particular case ⍳ and / map like "traditional programming langauge" funcitons?

it's going to be difficult for me to chat about all these glyphs unless i setup mapping on my keyboard

Adám

@0xACE For the chat, you could use abrudz.github.io/lb/apl

dzaima

6:04 PM

@0xACE they're not pervasive, aka scalar functions. Which you could interpret as them being more "traditional"

0xACE

7:04 PM

Geez, that was fun @Adam, great initiative. Thank you :D

Adám

@0xACE You're very welcome.

0xACE

7:19 PM

btw added '(d=>{let e=d.createElement("script");e.src="https://abrudz.github.io/lb/lb.js";d.body.appendChild(e)})(document)' as a grease monkey script to this chat. but now i can't type regular backticks at all, any ideas Adam (didnt want to highlight you as it's not that important)

dzaima

@0xACE does tapping ` twice not work?

0xACE

(⋄=2×/'backtick')
1

it produces a ⋄

dzaima

oh right yeah

0xACE

in dyalog it's 3 backticks and with the scrip its 2

dzaima

backtick+space works apparently

0xACE

7:22 PM

ah, nice catch, ty

another question, has anyone figured out rendering apl glyphs in the terminal? i bad with font config in linux

dzaima

@0xACE what terminal?

0xACE

im using st.suckless.org i have patched glyphs support for it but i dont think my font config is working correctly

they dont work in urxvt nor xterm for me... so it's likely a font config problem

dzaima

i'm using gnome-terminal, where the default font (dejavu sans mono) works fine

0xACE

oh i could try it out on one of the machines that are not setup as nerdly as mine and see if it's my config that way, good idea, thanks :D

Adám

7:58 PM

APL Campfire ft Ron Murray

►

goof

9:16 PM

@0xACE Works for me ootb on st (latest master) with all defaults (font = "Liberation Mono..."). Also in xterm with DejaVu Sans Mono. I know some terminals try to automagically fall back to some other font if the glyph isn't found in the main one, so idk if these fonts are actually being used.

LANG and LC_CTYPE & co are en_US.UTF-8 btw, if that makes any difference.

0xACE

@goof yeah i think it's my font config that is messed up... font config on linux from scratch has always been a mystery... I have the same ENV variables...

user

10:32 PM

@Adám Btw, would you say it makes sense to retag APL questions tagged reduction fold or something else? As your old question says, it has no usage guidance and it's pretty ambiguous. What would you recommend?

(On SO, I mean)

Transcript for