The APL Orchard

« first day (810 days earlier) ← previous day next day → last day (1847 days later) »

1:09 AM

⎕←↑[0.5] 'Hip' 'Hop'

@TessellatingHeckler
HH
io
pp

TessellatingHeckler

Axis A HALF ??? x_x

TessellatingHeckler

1:56 AM

0 0 0 0 1 0 0
0 0 0 0 1 0 1
0 0 0 0 1 1 0
0 0 0 0 1 1 1

the operation I want to do is condense consecutive 1 down to a single 1, but this will make the rows different lengths, so I can't work out how to do it with reduce or similar as they want fixed amount of inputs and outputs

ngn

2:13 AM

@TessellatingHeckler 2⊥⍣¯1⊢15 20

⍣¯1 means inverse function (literally: apply the function -1 times)

⊢ is to separate the ¯1 from the argument

TessellatingHeckler

@ngn what; how does applying the inverse of decoding, end up being the right quantity of bits

ngn

@TessellatingHeckler magic :)

@TessellatingHeckler mix (↑) introduces a new axis. the number in square brackets shows where it should be inserted

TessellatingHeckler

@ngn 😩

as I saw in a tribute to John Scholes:

('<'∘,⍣¯1) (,∘'>'⍣¯1) '<ha>'
ha
(Join ‘<‘ to the front and ‘>’ to the end of the letters <ha>, negative one times).

ngn

@TessellatingHeckler i'm not sure if this can be solved elegantly. probably split (↓) into a vector of vectors and process each (¨) one separately

TessellatingHeckler

@ngn I've got as far as splitting it into a vector, solving one at all is proving difficult

I was trying pairwise equality 2 =/ and logical operations

and had something which almost worked but left off the far right 1

ngn

2:22 AM

@TessellatingHeckler do you want to collapse only consecutive 1s (and not consecutive 0s)?

TessellatingHeckler

yes

they're going to merge back into strings, where the 0 is taken from the original text, and 1 is asterisk'd out

I want "t*" instead of "t***" because repeated wildcards don't do anything

ngn

@TessellatingHeckler something like this should work: {⍵/⍨~2∧/0,⍵}¨↓

TessellatingHeckler

@ngn that looks close to things I tried, and considered catenating a 0 placeholder to one side or another .. but what is commute doing in there?

ngn

@TessellatingHeckler ⍵/⍨something means something/⍵

it's a compress operation

TessellatingHeckler

@ngn that works so well it instantly shows me how it's the wrong thing for me to be doing. Now the 1 will also have moved so I can't use their positions as indexes into the text. Ooops

really seems like about 50% of APL's functions all do the same thing as each other

ngn

2:31 AM

@TessellatingHeckler there is some overlap but not as much as you think

@TessellatingHeckler i'm not sure i understand. do you want to convert the boolean mask to a list of indices?

TessellatingHeckler

(⍳5)[⍸1 0 1 0 1]

1 0 1 0 1/⍳5

ngn

yeah, those should be equivalent

TessellatingHeckler

@ngn I did convert them with iota underbar and merged my two matrices originally

now I can't change the size of one of them first, then merge them

should have thought of thatttt

DOMAIN ERROR
{⍵/⍨~2∧/0,⍵}'t***'
∧

so that would be the boolean matrix compress of omega

but I guess it's logical ^ which is not valid for characters

so maybe if I use = and instead of a prefix 0, a prefix character

⎕←{⍵/⍨~2=/'a',⍵}'t***'

Dyalog APL

@TessellatingHeckler
t*

TessellatingHeckler

YES! :D

ngn

2:36 AM

@TessellatingHeckler in this example you had booleans, not characters

using = doesn't sound right...

⎕←{⍵/⍨~2=/'a',⍵}'tt***'

Dyalog APL

@ngn
t*

ngn

you don't want to collapse the ts, do you?

⎕←{⍵/⍨~2∧/0,'*'=⍵}'tt***'

Dyalog APL

@ngn
tt*

TessellatingHeckler

@ngn yes in that example I had booleans not characters. Now I realise I can't do this removal of consecutive things until after I convert everything to characters, because if I do it first, I lost the information which tells me which characters to take. So instead, I need to convert to characters first, then remove consecutive asterisks

@ngn oh no, I don't, good point

I generate booleans for all combinations of asterisk positions, by counting 1..(2^length-1), in binary, and then overlay the asterisks on many copies of the original text

ngn

@TessellatingHeckler are you solving project euler? this reminds me of a problem there

TessellatingHeckler

2:43 AM

@ngn No, but I am solving a golf / challenge; it's not that I need an answer, but that I want to work through something in APL and try to solve (as much as I can) myself

TessellatingHeckler

2:59 AM

*et*pe
*et*y*
*et*y*e
*et*yp*
*et*ype
*etT*
...

noice

I think that's as far as I'm going to be able to get with this one; @ngn thanks for your help and explanations :)

ngn

@TessellatingHeckler my pleasure

TessellatingHeckler

whenever I'm quickly reaching for a dfn/each like {⍺..⍵}¨... that feels like it's not idiomatic APL and I should avoid it if possible; is there anything to that feeling?

ngn

Nov 6 '18 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.

TessellatingHeckler

haha, ok; (you found that quickly)

2 hours later…

Adám

4:59 AM

@TessellatingHeckler @dzaima @ngn Iverson deliberately made it so because base conversion is matrix multiplication by the base powers and matrix multiplication traditionally "transposes" its left argument:

⍞←(2*⌽⍳4) +.× 2 2 2 2⊤12 10 ⊣ ⎕IO←0

Dyalog APL

@Adám 12 10

Adám

@ngn On tryapl 1 ⋄ '' ⋄ 2 will give you an extra newline between 1 and 2. For the bot (and TIO, because that's what it uses) you need ⎕← in front of everything yo want to print. So ⎕←1 ⋄ ⎕←'' ⋄ ⎕←2 gives you an extra newline.

Adám

5:17 AM

@TessellatingHeckler Create an new axis of length ⌈/⍴¨'Hip' 'Hop' (3) "before axis 1" where "axis 1" is length ⍴'Hip' 'Hop' (2), so the result will have shape 3 2.

@TessellatingHeckler That is by definition ⍸1 0 1 0 1

3 hours later…

Adám

8:22 AM

@dzaima Can I get the exact hardware and OS you used for these tests? Also which compiler do you use for dzaima/APL?

dzaima

@Adám linux mint x64, some oldish i3 processor, java 10 with the ./build in the repository. can't give more detailed info rn, not home.

3 hours later…

Adám

11:16 AM

@dzaima I reported it first thing this morning and now 18.0 matches dzaima/APL for many of these. We're looking into the rest.

3 hours later…

dzaima

2:31 PM

this was my test setup - ⎕ex-ed in my APL, and the output executed in Dyalog:
https://tio.run/##hVLLattAFN3rKwZtxg/VlpSkDSFxMXjThWiQvQuhKHJaDG1iGlNohXalGCcTugnurtBsjAl00ZqAl/afzI@4594ZKaUEOouZc8899zUzaXL2IbnYbNqR/vI18J0uDpnl@vrmxUtAX@jLzyLLE@Dnoc9Lq98FffIYjdjhEI5d6YxSASC8ZkWC7b8VjUAKGdZD6QicWn1P5MOZ4AztWdgnONnkgBrTNRbVtng3DMlqVrSeMreeopZZZFiOhGSwss5k/S@dZUhVN6Jkn4zkgPf9xg71M54n3NalleirCbkbO/pqzJSQgZ7Cy0NasMWA9Kuf1gsUlsj6q47jXAzNvWVaLT04Pa0WeTOTWAQ9AvlqNkqhdXCx9GZCPorM0uq668T8wrzW91rN2hE91Wjw7nQ1E8hlgtxOzLVdj6u7AFxdq5vVLC7rdYsaPWwZcDI4Co/zsnQvMhOk2Hta3fHPSM/7p7CRj81XFOrryS3aIxdG0uoXqppvpJbcZ…

@Adám so it was actually something fixable? now i'm more intrigued

Adám

@dzaima We naively did the singleton extension and then log, instead of doing ln on the arguments and then the computation based on that. Furthermore, Java used to use the library C functions that we currently use, but they recently switched to assembler code supplied by Intel. We will have a look at using that too, or at least start special-casing log₂ and log₁₀ as the C library has dedicated functions for those.

dzaima

woah those arc-trigs
https://tio.run/##xZjNahRBEMfveYo570LZVd3VHxD1WWYTFEGiKAh6Eg8STCI5RNaD4MWTCLkseMkp@ybzIrHmq7uDMzlOhyVMd8jOj6r6/6u6j@qTd/Xbu7u6@XxJ66ekup/mYkfqYDOxd9BcnTdXH//7/JnfnPzTA/v951Le9On1m1fPq0oejt9XVfdw1C4@VMNqv320/1uNK/mPH1NfdTO/eTPz9psH2X61bM3Xb8cvK0B5rFApoPYhLnRLJ2gIEjq9OBytqY@J/FagRrRu0dK8OHlWKGzUXPysRzaPXdiedAtC8KpLalVpcGRCcThOcAq8jnABiReHqwUuZtWbrvIiHPMAZ0AF9EXg6qzkMJac9oChU4OWiPqAtgjcZoQzmEVO4MyYVpGqDcvDDYHr4SyBi3AYwHGsOXRMywtiFZOqGXxKqhOBJotj7ZZP6iq3OE4Wp8GaAc0Bea0LoOkRDRWEFLUAiAOaBvYeC6ClhDrQuUqNSgmlAkKoV1GjrWfkaBSydsrLR22VFGp0104HhTJ0omwVKsp1avmo0X6b@67JdcB2CJsX36UCEt…

my apl would have the advantage of all numbers being doubles already always, but it's not that big of an advantage

Adám

2:55 PM

@dzaima Me too:

      ]runtime -c "¯1○a" "(⊂'NaN')⊣@(=⍨)a"

  ¯1○a            → 6.5E¯1 |   0% ⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕
* (⊂'NaN')⊣@(=⍨)a → 2.0E¯2 | -98% ⎕

Dyalog APL