the k tree - 2021-08-03

« first day (879 days earlier) ← previous day next day → last day (656 days later) »

Bubbler

00:27

@ngn I only got to 70. Going down to 64 looks hard

Bubbler

01:25

66

@ngn I give up. I simply don't see a shorter way to create the five matrices, or to generate strings out of them, or whatever

JohnE

01:49

would an appropriate-based decode work?

would certainly make for an elegant formulation, but hard to say if it'd be compact

Razetime

@JohnE yeah this is what i was suggesting earlier

but ngn said it wasn't part of his solution

JohnE

oh, and curses- it's a variable upscale

wouldn't work as well, then

Razetime

yeah tis a different nxn block for each thing

1 hour later…

coltim

02:57

@Bubbler got it down to 60

Bubbler

03:09

Huh, TIL $<string> stringifies each char

Also nice use of boolean negation

@coltim Is # (in y#w!...) a reversed @ of some sort...?

coltim

@Bubbler yes; with dictionaries # takes a list of keys and returns those items (keys + values)

(may be worth noting that keys _ dict will delete those keys, returning a trimmed down dictionary)

Bubbler

03:25

And the keys are discarded on ,/, nice

Bubbler

03:39

Got 58 based on that

Bubbler

03:59

An interesting snippet: x#&0 2\x for an odd positive integer x gives 0 0 ... 1 ... 0 0 of length x with the single 1 in the middle

though it doesn't help much in golfing this specific problem

coltim

04:32

56

coltim

05:18

ergh, feels like there is some extractable structure to ~(+>\|=x;|=x;=x),+:\x#,i=|i:!x but it’s eluding me (something like x{y x}\(=:;|:;+>\) is cool, but takes more bytes)

A username

Can you use the current character instead of having a predefined string?

Bubbler

Maybe, ovs's solution has ord(c)%23%5

A username

That was clever

coltim

@Bubbler i think something like ` 0:"\n"/f[5;s] works for outputting

Bubbler

05:33

Nice

A username

Is there a way to stringify a char without the double quotes from $?

Bubbler

@coltim Golfed: {}` 0:'f[5;s]

coltim

(ngn/k parses `0 as the symbol form of “0” but that 0: overload needs to take ` as its left arg. to be honest i have never actually understood k/q’s i/o)

Bubbler

@Ausername Those double quotes are part of prettyprinting

coltim

@Bubbler case in point :P (how does an empty lambda do anything here?? could you put a semicolon at the end of it instead? why is any of this different than 0N!? mysteries)

A username

05:38

Then why does this happen?

Bubbler

@coltim I use {} to suppress prettyprinting the thing already printed

@Ausername Oh... that's a quirk of oK. It doesn't in ngn/k

Maybe you can replace $ with , (wrap in a singleton list)?

A username

Ok

I'll switch to ngn/k then

coltim

@Bubbler i think $: string in some k’s leaves things that are already strings (or at least lists of strings) untouched

@Bubbler hmm maybe some of the output helpers I’ve seen in J golfs will begin to make more sense to me

Bubbler

@Ausername (also every K on TIO is somewhat outdated. You'll need to use respective online interpreters to use the newest version)

A username

ngn/k has weird typing stuff.

Ok

Is there anything I can run K5 on aside from oK?

Bubbler

05:48

oK is actually close enough to K6 I think

Exact differences between oK and ngn/k are not very well documented but we can help you sort out the quirks

(btw, turns out that oK's $ does some kind of uneval, in that it wraps strings in quotes and escapes backslashes and double quotes)

A username

Ok, so this works in oK, aside from the quirk with ", but errors in ngn/k

ngn.bitbucket.io/k/…

Bubbler

First error is that you don't need to split with ";"\ because s is already a list of strings (list of lines)

A username

But then...

'val
..'!k};$[92=j;a[];47=j;|:'a[];45=j;h v,j;95=j;{k#" _"@x=k-1}'!k;v=j;(k;k)#v;+h v,j]}
                                                                ^

Razetime

btw why is a a function?

A username

Because it has two separate usecases

Can probably be golfed

Razetime

06:00

why's it a function instead of an array

coltim

@Ausername k and v need to be global variables to be accessed within the nested functions; can do k::y;v::" ";

Bubbler

Uhh... it doesn't look easy to fix, because ngn/k doesn't seem to recognize nested scopes (everything is local or global)

@coltim Oh, didn't know that

Razetime

@coltim whaaaaaat

coltim

although yeh, no closures in ngn/k (oK may be a bit of an exception k-wise; most don’t and rely on projecting/currying args)

A username

I'ma stay in oK for the time being.

Bubbler

06:04

Yeah, probably better to stick with one impl when you already have a (near-)working solution

A username

It is working, just $ working would save a byte

Razetime

Jun 23 at 9:12, by ktye

i don't like io verbs. what about:
c:<`file (monadic/read)
`file<c (dyadic/write)

if implement a K, it'll probably have this

Bubbler

@Ausername So j in there is always a single char?

A username

Yeah

Bubbler

I guess you can't save a byte at that particular point

original code had v,j instead of $j, right?

Though there's a high chance you can use "pad" to save bytes on appending/prepending some spaces

A username

06:13

@Bubbler There's a pad builtin? Cool

Bubbler

Yes, ctrl+f "pad" on oK manual

ngn

06:27

@coltim @Bubbler very nice! you can save a byte by turning w into an argument and making a projection like this

here's mine - 51 bytes: {,/,'/'`c$32|x*(a@<a;+a;|=y;=y;a=:|a:y#'!y)8!19!-x}

Bubbler

@ngn Oh wow, I definitely saw that but never thought of golfing with it

A username

I do not understand any of this

Bubbler

Also I knew it must be golfier to use remainder trick somehow

ngn

@Ausername i'm sure you understand { } and ,/. do you understand ,'/'?

A username

,/ is flat, {} wraps in a function, but no clue what that is

I presume 8!19! is to do with some remainder trick

ngn

06:36

@Ausername ,'/' is "catenate each over each". if you have a matrix of matrices, it joins the rows horizontally

A username

Oh that's nice

ngn

but it's probably better to start from the right

A username

What does a@<a do?

ngn

@Ausername sort

a there is the matrix corresponding to "-"

all 0s except for the middle row which is all 1s

A username

+a is for | right?

ngn

06:38

so a@<a brings the 1s row to the bottom and we get a matrix for "_"

A username

Cool

I think I understand it now.

ngn

@Ausername +a means "flip" over the main diagonal (a.k.a. "transpose")

A username

Bye for now

@ngn Yeah

ngn

@Ausername right

@Ausername o/

A username

06:59

I'm back

So basically how I think it works (correct me if I'm wrong) is:

Build a list of binary matrices, modifying the previous each time

Map each character to that with a modulo trick

Replace 0s with spaces and 1s with the relevant character

And transpose, concatenate each row and flatten.

Is this right?

ngn

@Ausername yes. so, you understand all of it :)

you might be wondering how " " is handled. 8!19!-x maps it to 6 which is out of bounds for the provided list of matrices. but k supports out of bounds indexing - in this case it will return a matrix of "nulls" (0N which is the smallest representable signed int). 0N*" " happens to be 0 which suits us just fine :)

the solution might be even shorter in k9 where char arithmetic returns chars and sort is ^x instead of x@<x

ktye

07:49

@ngn did you ever try test coverage? i guess it's hard in non-line oriented programming style. i started doing that, and found many bugs immediately.

ktye

08:12

@dzaima do you have range types in bqn? e.g. represent a+b*!c with 3 ints only. k7 had a+!b ranges. i could think of many operations that preserve the compact form. but again, is it worth it? without doubt, all these optimizations add to the code base..

chrispsn

@ktye i want to know a lot more about your implementations

and what has changed between iterations and what you're learned

a write-up would be really interesting

@coltim it would be nice if there was a composition verb for this purpose (compose\fnList)

@coltim something that kinda works but again, is longer - stringify, merge, execute: (.',\$(1+;2*;3-))@\:3

ktye

@chrispsn good idea, i think i can do that.

chrispsn

@ktye :D

2 hours later…

ngn

10:21

@ktye no and i'm not sure if i'll ever get to that

dzaima

@ktye Not yet at least. Adding them would mean everything reading integer arrays has to check for them, which, as I found out, makes things reasonably slower. I'd rather have an optimization pass over the byteceode, searching for such patterns and replacing them with better code

(fwiw, CBQN currently has only i32/c32/f64 typed arrays, so adding i8/i16/c8/c16 (and bit booleans) is a priority over fancy optimized types)

dzaima

10:35

@dzaima the BQN solution

13 hours later…

Bubbler

23:18

CMC: Build me a room in K

coltim

@Bubbler I was playing around with this but it fails on 2’s and I gave up

Bubbler

I'm pretty sure the best way is to generate a flat version using base (or !x with x a vector)

dzaima

23:35

@dzaima (note that marshall wants thunks at some point, so who knows how things will look in the future)

« first day (879 days earlier) ← previous day next day → last day (656 days later) »

Transcript for

Aug '213

the k tree

app.element.io/#/room/#ngnk:matrix.org

join this room
about this room

00:00

06:00

12:00

18:00

all times are UTC

site design / logo © 2025 Stack Exchange Inc; legal

mobile