i might be missing something obvious but is there a straightforward k method of doing apl's partition ⊆? (following @code_report's video)

'group' is closest

@chrispsn cut?

except it takes a boolean mask instead of list of indices

oh, i was thinking of ⊂ .. ignore me

maybe something like >': ?

(&0N>':x)_y

it's kinda fiddly

@chrispsn @JohnE i've just watched (a few minutes from) the video - the guy there is using ⊆'s special case for 0 to remove non-digits and split by them. i don't think k has a short equivalent of that.

doesn't quite handle all input - a wip. k9. {c:":"; #=.'(c>)#'x@=+\" ">':c<x} link

@chrispsn i was thinking: we could replace non-digits with spaces, collapse runs of multiple spaces, and eval

d:`c$"0"+!10
#?.(" "/"  "\)/d@d?x

@ngn i can't get this working in ngn/k

@ngn {#?.(#')#":"\":"&x} in k9 (char&char => char)

not sure why but #?.(#')#":"\":"& didn't work

@chrispsn i think the web version is too old and has a bug. just a mo..

strange. something's wrong with outdexing in wasm.

i think i know what. i'll fix after lunch.

@chrispsn should work now

@ngn local link?

should work now v2.0 :)

0N is always 64bit in my impl and it was lost in wasm because sometimes i use size_t for indices

define "should"..

"what's your opinion about.."

for my own programming, i prefer stupid straightforward primitives. i can deal with 0N and 0n explicitly if i have to.

data scientists may not think this way

@chrispsn what does "log W" mean there?

this formula?

@ngn yes, and 'WASM/Mobile'

There's more but it's best explained by @kelas

@ngn strictly for consistency with other k's (strictly), what about allowing any amount of whitespace between numbers in the vector/strand notation? going off TiO, both kona and oK have that behavior

@chrispsn to translate this a couple tweaks are necessary: #?.'(#:)#":"\`c$":"&

@ktye the blurb here makes it sound like the "ignore nulls" part is to align with SQL o_0. How do other libs/languages handle nulls in aggregations?

@coltim julia wants you to be explicit: docs.julialang.org/en/v1/manual/missing/…

@coltim i can't decide. anybody else with a strong opinion?

k9 uses strictly 1 space

IMO strictly one space

@ktye thinking about it more, I don't know how often I've taken advantage of nulls being ignored. I know I've been bitten by the "running an aggregation function strictly on a list of nulls", which caused some headaches when using min/max (introducing infinities in weird places)

re whitespace in stranding: I don't know if it'll ever really come up in practice in people's actual code (I don't see any reason why someone would use more than one space). on one hand allowing arbitrary whitespace is consistent with the rest of the language, but I guess vector/strand notation is somewhat of a special form. it would have uses golfing, as in this example

the main reason for people wanting this is alignment

hmm, good point! could even result in "pretty-printed" output still being evaluate-able

@ngn @coltim would spaces in symbol lists be a similar case?

@ktye and it looks like numpy has a separate nansum() function to treat NaNs as 0's

@chrispsn maybe. i'm not sure.

@ngn or copy paste from somewhere else

#?.(97>)#'(30<-':)^ (or similar) could work too

or #?.";"&(30<-':)^

@chrispsn very clever!

hmm doesn't work on empty string

how does it work? what is f^?

f^ seems to be just {(&f x)^x} (like good old cut)

yeah, ^ is cut

is there an option to enable round-tripping?

not sure

my last answer above cuts where the gap is too big for it to be anything other than an int -> char transition, then turns all chars into semicolons, then evaluates (plain semicolons are empty statements so only the last one gets picked up)

ah..

yeah, this is nice

but it doesn't work where there's trailing chars or an empty string

my eyes hurt after a few minutes staring at kelas' web repl :)

. on a list of strings penetrates the list?!

yep, in k9

@chrispsn is there no eval/apply anymore?

isn't . still eval?

oh, as in dyadic?

previously .(x;y) was the same as x@y

and .(x;y;z) was x[y;z]

oh! i have never used that

not sure

#?(`i=@')#.";"&(30<-':)^  / deals with empties and trailing chars

@chrispsn you could also do ." "/$ instead of (`i=@')#

(handles trailing letters but fails for the empty string)

now this is strange: ?(0;"a") -> domain error

heh

and .,";" fails but .";;" is ok

i'll report it

@chrispsn Where?

Ah, that's basically what I was doing on one of my many implementations.

@Adám cool - thanks for having a look

doing a simple string cat-and-eval seems faster than the cut approach

{#?(`=@')_.";"&"(",x,")"}

@chrispsn is there still a way to parse? `p@ doesn't work

@ngn not at the moment

been playing around with some ideas for utility functions; came up with this for insert

was thinking of extending it to take a list of indices as well, which would be combine (e.g. apply some rank/depth-reducing function to the values at those indices, removing them from the input and inserting the new value at the first index)

@coltim that sounds very much like splice

@ngn yeh, it's definitely modeled after splice. Was there ever a "formal" version of it? I know oK has it, although I'm unsure of what I think about the "interval" aspect of it

@coltim "formal" - nothing ever is in k

prepend with a letter to ensure groups alternate [non-digits,digits], then crack at run breaks, refold, drop words and evaluate

oops, I mean

?.(`c$.[255#32;d:48+!10;::;d])@