The APL Orchard

nathan rogers

4:38 AM

@Adám this is for spheres

nathan rogers

4:48 AM

@Adám by finger count, new solution is around 5 seconds, and old solution around 18

tio.run/##7VbNaxtHFL/…

The indmin function isn't returning the results I want, as I want to return specific colors base on what was hit, though that's probably already solved I just need to map the results some new way

nathan rogers

5:13 AM

actually, much easier than I thoughtcolors[(Pcs Scs 0)[l]]

ngn

6:27 AM

@arcfide what do you think of ,\0 1 2?

nathan rogers

6:51 AM

@ngn how do you drop a dimension?

I have a shape 1 2 8 6 and I want to drop the 1

so that I have a shape 2 8 6

ngn

@nathanrogers reduction deletes a dimension. you could do ⊢⌿ for instance

nathan rogers

@ngn i have a 2 8 6 and an 8 6

I would like to make a 3 8 6

ngn

@nathanrogers ⍺⍪↑,⊂⍵ ⍝ there may be a better way

Adám

7:20 AM

@nathanrogers A286⍪A86

nathan rogers

when I have an 8 6 and an 8 6 it doesn't work

it creates a 16 6

Adám

@nathanrogers True, but that wasn't your question.

nathan rogers

ok, how do I extend joining 2 8x6 matrices such that I wind up with an additional leading dimension?

8x6 and 8x6 makes 2 8 6, a 2 8 6 and 2 8 6 makes 4 8 6

Adám

@nathanrogers For that you need a,[¯0.5]b or ↑a b

@nathanrogers Those two operations are fundamentally different, as the first adds an axis, the second not.

@nathanrogers If you want a general solution (specific, actually) that handles both rank-2 and rank-3 args, use {(⍺⍴⍨¯3↑1,⍴⍺)⍪(⍵⍴⍨¯3↑1,⍴⍵)}

{⍵⍴⍨¯3↑1,⍴⍵}{(⍺⍺ ⍺)⍪(⍺⍺ ⍵)} is an alternative way to write that without code duplication, and in my extended APL, ⍪⍥{⍵⍴⍨¯3↑1,⍴⍵}.

nathan rogers

Adám

7:30 AM

@nathanrogers I can't see images on the train.

nathan rogers

:(

imgur.com/d5wRtAr

Adám

@nathanrogers I mean, the train blocks imgur. I'll just look later when I get to the office.

nathan rogers

oh that's right other countries don't celebrate thanksgiving

I give thanks for the extra days off

(2 3 4⍴⍳6){(⍺⍴⍨¯3↑1,⍴⍺)⍪(⍵⍴⍨¯3↑1,⍴⍵)}(2 3⍴⍳6) or reverse args

I'm not getting something

(⍳2),[¯0.5](2 2⍴⍳4)

also not waht i expected

Adám

@nathanrogers The planes don't match; 3 4 vs 2 3!

nathan rogers

oi derp

lol

thanks

Adám

7:35 AM

@nathanrogers Here you are pairing a rank-1 with a rank-2 and adding a new axis. Expect mess.

@nathanrogers ,[k] requires otherwise conformity. ↑a b will pad with the prototype of a.

Adám

10:27 AM

@nathanrogers Remove ¨ from indmin[3]: ds←inf@(≤∘0)⍵ for a significant speed-up. If you want even more speed, refactor your formulas to only use real numbers, as complex arithmetic is slow.

Adám

12:19 PM

@dzaima I just added ⍮ and noticed a new idiomatic function: ↑⍤⍮ "rank-up". Monadically, it increases the rank by 1. Dyadically, it pairs the arguments along a new leading axis, e.g. two vectors become a matrix, and two matrices become a 3D array. The args have their ranks and shapes aligned appropriately. :-D

Adám

2:07 PM

CMC: 0 0⍴' '

Quintec

2:21 PM

7 bytes :P

Adám

2:32 PM

@Quintec Uh, the CMC spec is 7 ⍨ I've got 4. Just discovered it today.

@Quintec Also, there is one obvious byte to save.

ngn

2:45 PM

@Adám i got 5: 0/⍪'' still trying...

Adám

@ngn Yup, I knew someone would find that one, but this one is… unusual. Think out of the box!

Cows quack

I get 3 for 0 0⍴⍬, but still trying for ''

H.PWiz

@Adám Another 5 is ⊢⌸⍨''

Cows quack

and another 5-byter ↑⊆⍨''

ngn

3:07 PM

@Adám i found it

Adám

@ngn ⎕NL⍬

ngn

@Adám yes :)

{0::⋄(0 0⍴'')≡⍎∊'⎕'⍵'⍬':⎕←⍵}¨,∘.,⍨⎕a

i guessed the general form of the solution and tried all ⎕XY combinations

J. Sallé

Hey guys, is it possible to set a recursive Dfn to sleep for some time before calling the recursion?

Adám

@ngn Ah yes, you don't know how to ask the interpreter which 2-letter ⎕-names it knows…

@J.Sallé Sure, ⍺∇⍵⊣⎕DL 3

J. Sallé

3:23 PM

@Adám oh nice that's what I needed. What does the DL stand for? Also, is 3 the time it'd sleep for?

Adám

@J.Sallé DeLay, and yes, time in seconds. It returns the actual elapsed time shyly.

ngn

@Adám how?

J. Sallé

Great, I think that's gonna work then

Adám

@ngn Secret feature ;-) The interpreter can do a lot of stuff we don't tell about.

J. Sallé

@Adám another thing, is there a way to insert a linefeed in a string other than ⎕TC (Mastering Dyalog APL says that's obsolete but I can't find an alternative)?

dzaima

3:32 PM

@J.Sallé ⎕ucs 10 is what I've seen used everywhere

J. Sallé

@dzaima Oh nice, thanks!

dzaima

usually as nl←⎕UCS 10⋄... so you don't need to spam parenthesis

J. Sallé

indeed

Adám

Maybe even nl←{⍺←⊢ ⋄ ⍺,(⎕UCS 10),⍵} so you can write 'Abc'nl'def'nl'ghi.'

J. Sallé

I'll only need it once, so just doing ,(⎕UCS 10), works fine for me

J. Sallé

4:51 PM

@Adám can I have a while-like statement in a Dfn? Basically I need my function to call itself recursively while ⍵≤5, but I also need it to output some text, so the usual cond:true⋄false statement doesn't support that afaik

or should I just make an inner function to do that?

Adám

@J.Sallé Can you reverse the condition to allow you multiple statements if true?

J. Sallé

@Adám I think I can. You mean do something like ⍵≥5 instead of ≤?

It'll add a byte or two but it's not a big deal

Adám

@J.Sallé No, ⍵>5

J. Sallé

I need that 5 though so I'll need either ⎕IO←0 or ⍵>6

Adám

@J.Sallé Alternatively, you can use ⊣ instead of ⋄ (and reverse order of statements)

J. Sallé

5:01 PM

That might also work, lemme try it

Adám

@J.Sallé I don't get it.

J. Sallé

@Adám I'll explain, just a sec

Okay so, I'm trying to solve this challenge. Basically, I need to print a fixed string every 2 seconds 5 times, and each time the string needs to be followed by a linefeed and a (randomly selected) . or !.

So basically it'll print string\n for tick 0, then string\n., then string\n.! and so on

Currently I've got {⍵≤5:'Sending 5, 100-byte ICMP Echos, timeout is 2 seconds:',(⎕UCS 10),'.!'[?⍵⍴2]⋄∇(⍵+1)} but it'll obviously only print once

All I need is to print that 6 times (with ⍵=1-⍨⍳6)

@Adám ↑

Adám

5:21 PM

@J.Sallé Sorry, got distracted.

J. Sallé

No problem, this is definitely less important than whatever you were doing hahahah

Adám

@J.Sallé Doesn't this work?:

⋄ {⎕←'Sending 5, 100-byte ICMP Echos, timeout is 2 seconds:',(⎕UCS 10),'.!'[?⍵⍴2]}¨⍳6

Dyalog APL

@Adám
Sending 5, 100-byte ICMP Echos, timeout is 2 seconds:
.
Sending 5, 100-byte ICMP Echos, timeout is 2 seconds:
.!
Sending 5, 100-byte ICMP Echos, timeout is 2 seconds:
!!!
Sending 5, 100-byte ICMP Echos, timeout is 2 seconds:
!.!.
Sending 5, 100-byte ICMP Echos, timeout is 2 seconds:
.!!..
Sending 5, 100-byte ICMP Echos, timeout is 2 seconds:
!.!!.!

J. Sallé

@Adám nope, there has to be a delay of 2 seconds between outputs (and something else I just noted, the .! must stay in the same order)

that second part will be a pain

Adám

@J.Sallé Sure, but you can put in a ⎕DL inside the dfn.

J. Sallé

5:26 PM

also the first message has nothing after the linefeed

Adám

@J.Sallé What order do you mean?

@J.Sallé ⎕IO←0

@J.Sallé For code golf, you do want the deprecated ,⎕TC[2], instead of (⎕UCS 10) or even use ⎕ML←3 and use ,2↓⎕TC,

J. Sallé

@Adám I was about to mention that. If the second message has a . after the linefeed, the third needs to have that same dot in that position

Adám

Getting off train. Later!

J. Sallé

No problem! I'll probably finish this up tomorrow

nathan rogers

7:37 PM

holy shit yeah that is a significant speed up! @Adám

Adám

Nov 6 at 6:17, by dzaima

@nathanrogers a simple way to score the speed of an APL program is count the number of ¨. The more, the worse it'll be.

nathan rogers

But I mean, from 5 seconds to almost instant

That's crazy

Adám

@nathanrogers Hey, can you count how many ¨ were in your original code, and how many are there now?

nathan rogers

@Adám unfortunately as far as complex arithmetic, that's only due to the quadratic formula. The sphere detection formula returns a complex number if it doesn't intersect. Unless there's an alternative sphere intersection formula, I don't know how to get rid of that

Adám

@nathanrogers OK, but anyway, it is "fast enough" now to be workable, right?

nathan rogers

7:41 PM

yeah, it's almost instant

even at higher resolutions

with multiple objects

dzaima

@nathanrogers the power of APL when you use it well. :D

nathan rogers

There weren't many ¨'s there, but there was a lot going on in the ∘. function, now the ∘. is just a simple multiplication, and the new data structure makes it really fast

I had 3 in the original ¨ and 3 ¨ now, ¨ plane ¨sphere and ¨width and height

I did a good job of keeping @dzaima in mind when making my first go :P

but I didn't know about "arrays", and certainly hadn't used anything higher dimensional to solve any problems

Adám

¨-)

nathan rogers

so this was a crazy good learning experience

Adám

@nathanrogers Good. See? You are getting there.

nathan rogers

7:45 PM

4000 by 2000 resolution took 2 seconds

that's crazy

Adám

@nathanrogers How long does it take in C++?

nathan rogers

let me see

dzaima

@nathanrogers I still haven't begun touching 3D+ arrays, so that's one thing you're more experienced at than me :p

nathan rogers

5 seconds

Adám

@dzaima We have customers that threaten to give APL up if we don't increase maxrank.

dzaima

7:47 PM

@Adám oh god

Adám

@nathanrogers Heh, interpreted APL beats compiled C++!

nathan rogers

this is really good news. uncompiled, this is really fast

Adám

@nathanrogers And that C++ code is well written, by somebody who knows their stuff?

nathan rogers

I was using this as sort of a test bed to see if graphical APL would be feasible, and I guess now I know that it is

He's worked on the Witness with johnathan blow, so I'm assuming yes?

I've watched a lot of his content, and he really knows C and the Win32 API for graphical programming inside and out.

He didn't optimize anything, so maybe optimized it would be faster and I'd have to actually time things for real

Adám

@nathanrogers Well, that kind-of proves my oft-repeated claim that APL often can beat carefully hand-written C/C++ despite APL being written in C/C++.

@nathanrogers I wonder if co-dfns would gain you even more speed.

nathan rogers

7:50 PM

I am writing @arcfide right now about this version to see if it could be compiled

I'm using codfns for the display, but I don't know how to namespace, or compile things

at 4k resolution, the new one takes 2-3 seconds, the old one takes 45 seconds

incredible

@Adám who is the guy you spoke to about my original version?

Adám

@nathanrogers Marshall

@nathanrogers how to namespace?

nathan rogers

yes

I don't understand Dyalog project structure, and namespaces are really wierd. I've tried to make namespaces and then load them, but then it says "not a namesapce"

so I really don't get it

Adám

@nathanrogers You may want to have a look at Lesson 16.

dzaima

@Adám 30s of looking tells me that doesn't talk much about loading/saving them

Adám

@nathanrogers To save and load them, look at the end of Lesson 28.

nathan rogers

7:59 PM

@Adám did you catch my question from last night about how to drop specific ranks?

Adám

@nathanrogers Uh, no, I think I missed it. Can you link me?

dzaima

@nathanrogers didn't ngn already answer with ⊢/ (or ⊢/[dim])?

nathan rogers

13 hours ago, by nathan rogers

@ngn how do you drop a dimension?

Just wondering if there was a more standard solution, as I seem to recall a discussion about a specific operator which dropped specific ranks

but that might have been a misunderstanding on my part

dzaima

@nathanrogers I don't think there's anything better than ⊢⌿

dropping a rank if it's exactly one is pretty specific (though I think a built-in type of thing that errors if it's not one would be helpful)

Adám

@nathanrogers I saw that, but ngn answered right. If you want to drop dimension k, just do ⊢/[k]

nathan rogers

8:02 PM

ah

there's something goofy going on with the spheres, because each sphere, if hit, will return 2 values

2 matrices, so get the min to get the closest between them

but then if multiple spheres intersect on the same point, I get back multiple lists

Adám

@nathanrogers Doesn't every ray that hits a sphere have an entry point and an exit point?

nathan rogers

so I was getting back n x 1 x h x w shapes

yes, but I onlly get the min of the sphere

so the closest hit point

say 3 spheres intersect on the same point, 3 1 h w, would be the shape

so I did ⊢⌿↑⍉

perhaps just ⊢⌿[1]

dzaima

@dzaima oh right something stupid like {÷}⌿ works (as typing ⎕signal is way too much work) :P

Adám

@dzaima That's actually pretty clever.

nathan rogers

mindex Pds {(⍺⍴⍨¯3↑1,⍴⍺)⍪(⍵⍴⍨¯3↑1,⍴⍵)} ⊢⌿↑⍉Sds

using your 3d specific append from yesterday

@dzaima I don't get it

Adám

8:09 PM

@dzaima Hey, you finally found a use for a niladic derived function! (Do you know how to make one?

dzaima

@Adám 2∘+∘3⌿? :D

Adám

@dzaima Save 2 bytes: 2∘3⌿ (!)

nathan rogers

???

dzaima

@Adám oh I thought that'd be a syntax error all the time, cuz i did 2 . 3 at first

Adám

@nathanrogers Never mind, we're being silly.

@dzaima . is special :-( 2⍣3/ works too and so too any normal dyadic op.

dzaima

8:12 PM

relatedly, a thing I'd really like in every language to be is a syntactical construct for "TODO" that errors upon getting reached

Adám

@dzaima Syntactic construct? We tend to write ∘∘∘ or ...

dzaima

@Adám ...⌿1 2 3⍴0 errors immediately though

Adám

Wat?:

⍞←0⍣0/1 2 3

Dyalog APL

@Adám 3

Adám

@dzaima Yes, as a I said, . is special (due to nss and ∘. and 2.3)

dzaima

8:14 PM

@Adám so does ∘∘∘⌿1 2 3⍴0

Adám

@dzaima Yes, because that isn't a valid derived fn.

dzaima

but yeah, that expression is a very special case of requiring erroring

Adám

@dzaima Sure, but the parser doesn't even manage to make sense of anything. 2∘3 is perfectly OK syntactically.

@DyalogAPL Ah, I get it; you don't care about the left operand if the right operand is 0.

dzaima

@Adám heh, makes sense why that'd happen but still feels strange :p

Adám

Conditional inline syntax error: 0⍣cond⊢

dzaima

8:18 PM

:D

Adám

However, I still think a bunch of operators which only take functions as left operand should be extended to treat an array left operand as A⊣⊢

Then you could conditionally replace a value by another with replacement⍣condition⊢normal

@dzaima Hah, I found a shorter (albeit even more obscure) one: 0&/

dzaima

@Adám heh, &/ was one of the first things I tried, forgetting that it's an operator

Adám

@dzaima This of course works with any monadic operator which doesn't take an array operand (that's all of them, no?)

dzaima

@Adám ⌶ is an exception

Adám

@dzaima Ah yes, it only takes a simple scalar positive int.

Transcript for