@Bubbler the argument that finally convinced me to implement rank-sensitive find (it took a long time) was that "find" should be (as close as possible to) an inverse of "at"

for instance, it's nice to be able to make a dictionary and index it like this:

 d:("key";"anotherkey")!("value";"anothervalue")
 d["key"]
"value"

instead of

 *d[,"key"]
"value"

Hmm

dict indexing is "find it in the domain and at it in the range", i.e. if d:k!v then d[x] is v@k?x

so it's nice for @ and ? to complement each other in cases like this

but (as i tried to argue above) "except" is a different matter

That somewhat makes sense for "find"

combined with the fact that the right arg of @ can be a multi-dimensional array of indices

And yes, "except" is unrelated to this matter

@Bubbler yes, or a ragged array, or a mixed-rank array..

But then how do you actually do a rank-sensitive find? Recursively dig through y until a match is found in x?

So both x and y are required to have some uniform rank?

@Bubbler only x is required (though i don't actually verify that, i just assume it.. for performance reasons)

rank (with assumption of uniformity) is defined as: rank(atom)=0, rank(list)=1+rank(first(list)), rank(dict)=rank(values(dict))

So it somewhat works if x is uniform and y is not too ragged

i think it should work whatever y is

i mean, in some cases it may error..

but you can do things like ("ab";"cd")?("ab";,"ab";,,,,"ab";())

Btw why not 0N instead of error?

good question

@Bubbler if ("ab";"cd")?"a" is 0N, the inverse ("ab";"cd")@0N is " " (two spaces)

i'm a little uncomfortable with "a" and " " having different ranks

Oh ok, I guess it's a sensible decision

johnearnest.github.io/ok/ike/ike.html?key=M7RwLwQ1 I got the required rice field pattern, but the first band isn't matching up quite right

might be a floating point inaccuracy

oh re. the identity matrix thing there's also a cute tacit version in dialects which have a rotate primitive, like k3: (-1!)\0=!:

it works because converge-scan halts when you revisit the initial value in the sequence

ah nice

worth noting that in some dialects (early k5, I think?) the result was the transpose of what oK and ngn/k produce now, for reasons which may be obvious given the above implementation

@JohnE sure, I'll add it in

which dialects do not have odometer

hmm. well, k2/3 don't have it

dunno about k4

in k3 it would be {x _vs!*/x}

in q it's {x vs til prd x}

and I don't see a mention of odometer in the kdb+ documentation, nor does ! appear to have it as an overload in k4

sure, I'll also add the k3 one

I don't have kona installed on this machine but the k3 one will probably work there too

@JohnE Try it online! works alright

@JohnE this is k6 right

haha well looks like kona gives you the transpose of the k3 version

the original version I gave was k6, yeah- works in oK and should work in ngn/k

I think if you dig around in the right places it's possible to find a real copy of k2 online. I have a personal license for k3 which I cannot redistribute, but I can always try stuff out for people

cool, added

so to clarify the k3 version I provided has the exact same behavior as the k6 version. Kona, however, appears to have different semantics (a bug?) for _vs

so kona gives you transposed output

@Razetime neat! i think the issue may be the denominator; what about this?

pretty trippy animated: johnearnest.github.io/ok/ike/ike.html?key=fTHn_Ivs

one general note: if you're drawing a single matrix as a bitmap, you can leave the position off entirely (like (;pal;image) ) and it'll automatically be centered on the display. This saves a few characters in this program and makes it easier to make stuff resizable.

also while oK doesn't have it, iKe adds a pow[x;y] function along with several other handy math functions

hi guys

welcome

hi guys

@JohnE canvas needs to be size of the args

@coltim idk how that works tbh

because the questions says R^s

oh, is this a golf problem

but removing the power somehow works

@JohnE yepyepyep

@MaximusSu welcome, new to K?

@JohnE using w and h will save a bunch of bytes

as well as just generally untangling all those nested lambdas in the original version

did not know pow existed in iKe

github.com/JohnEarnest/ok/tree/gh-pages/ike#built-in-functions a perlin noise generator, too

@JohnE quite maddening to forget that i'm using an array language

A.B.C.: Always Be Conforming

whatzat mean

conforming is what atomic operators do: line up scalars to vectors and vectors to vectors and so on

'll write that down

I'm also making an oblique reference to a famous scene in the film Glengarry Glen Ross: youtube.com/watch?v=w7y6EOGY87U

added to watchlist

wow this movie is stacked

it's one of the most crystalline representations of the mind of a salesman in cinema

I've seen whiplash for the musician, blackswan for the dancer's mind, birdman for the performance artist

I'll tell you when i watch it :thumbsup:

Wag The Dog might be an ideal example for PR

or you could just read edward bernays' Propaganda

probably the most societally-damaging book of the 20th century

"full on horror disguised as a buddy comedy"

that's a description

@Razetime @Bubbler this looks wrong: <syntaxhighlight lang="apl">

needs a highlighter for k

empty highlighter works well for k :)

mm doesn't sit right with me

but if it doesn't look correct I'll remove it

one example: ' delimits a string in apl but is an adverb in k

yep removed it

thanks

@JohnE that is super cool

@Razetime in case you didn't know: you can use 1-space indent instead of <pre> or <syntaxhighlight>

@ngn is there any way to convert a number to hexadecimal in ngn/k? `c$ only works if the value is outside the printable ASCII space

maybe passing that into a hex@?

@coltim "0123456789abcdef"@16\ ?

@coltim or using stuff from golf.k: (𝔻[],𝔸[])@16\

@coltim `hex@ is for strings (currently)

I see. it is for a golf, and I think the value I would want to output in hexadecimal would be <255, so maybe I can get away with hex@`c$ here

@ngn hmm, this works when run locally (returning !0) but throws a 'lmt on the wasm version

something's wrong with "where" on an empty int list in wasm: &!0

@coltim works now

@ngn that was fast, nice!