Vyxal

user

01:56

@AndrovT awesome, that makes life much easier

lyxal

02:37

!!/amilyxal

Vyxal Bot

@lyxal You are lyxal.

lyxal

@VyxalBot if you say so

The Thonnu

07:20

!!/amilyxal

Vyxal Bot

@TheThonnu You are not lyxal.

lyxal

07:32

Okay so I think I'll try removing the implicit conversion and see if that fixes anything later tonight

Because maybe something is being a bigint or something where it shouldn't be

But I think it might be something to do with the predictions function

Because it does the first loop just fine exactly the same and then starts deviating from the python results

Also, we might need a new name for vycoder

github.com/vycoder

Mfer really named like a compression algorithm

emanresu A

... why is that an issue?

It's not like we're creating a separate account

lyxal

I want our compression algorithm to come up when searching for it

Not someone else

lyxal

07:51

sigh

I found the bug

It's stupid

  .map((a, b) => distanceWeight(BigDecimal(i)) * (a + b))

That's the problematic line

out = [distance_weight(i)*x + y for x, y in zip(positions[i][lst[-i-1]], out)]

That's the original python

Turns out I forgot how operator precedence works

x * (y + z) != x * y + z

I have fixed this now

Haven't tested because mobile but should be equivalent now

Vyxal Bot

08:36

nayakrujul created branch anti-diagonal-patch in Vyxal/Vyxal

nayakrujul opened pull request #1721 (Make anti-diagonal (Þ) not error for an empty list) in Vyxal/Vyxal

Vyxal Bot

09:00

Lyxal approved #1721 (Make anti-diagonal (Þ) not error for an empty list): "I am clearly very good at programming. The best. Obviously. "

nayakrujul enqueued pull request #1721 (Make anti-diagonal (Þ) not error for an empty list) in Vyxal/Vyxal

nayakrujul merged pull request #1721 (Make anti-diagonal (Þ) not error for an empty list) in Vyxal/Vyxal

Lyxal deleted branch gh-readonly-queue/main/pr-1721-ef5bc19cc4b3aa92655ce3142649f296ff70e7ca in Vyxal/Vyxal

nayakrujul deleted branch anti-diagonal-patch in Vyxal/Vyxal

lyxal

09:20

Silly idea but what if vycoded programs were encoded several times?

Like take the output, vycode that and rinse and repeat

On mobile so I can't test, but I wonder if that would work

And by how much could multi layering shorten stuff

emanresu A

Have you tried zipping a zip file?

(as I understand it) that'd basically be changing it from common digraphs to common trigraphs

AndrovT

Compression relies on noticing and exploiting some pattern. After the first round of compression there is usually no pattern left or it is much harder to use after the compression than in the original compression algorithm.

emanresu A

09:37

^

Unrelated String

and repeated compression probably also makes it way less stable for novel inputs

since whatever patterns it might find in the compressed output aren't necessarily going to be anything more than flukes

lyxal

10:34

As I said, a silly idea :p

AndrovT

10:51

The most obvious way to improve the compression is to make the prediction model more accurate. One idea is to make Vycoder understand Vyxal syntax. For example ; can only be after one of λ ƛ ¨Z ⟑ ' µ ¨2 ¨3 ¨₂ ¨₃. Compressed literals should also be excluded from the compression.

Oh also digraphs should ideally be one symbol for Vycoder

lyxal

11:12

How would that work?

Extend the code page to contain a list of digraph characters and lex digraphs as one thing in the data?

AndrovT

Right now vycoder does roughly this:
Vyxal codepage -> numbers 0..255 -> encode

what you want is:
Vyxal code -> lexer -> numbers 0..n -> encode

lyxal

11:36

:+1:

I'll look into that once I get the scala working

lyxal

12:24

so turns out that while the precedence thing was an issue, it wasn't the only issue

it's still not matching

AndrovT

12:49

github.com/Vyxal/Vycoder-Scala/blob/… this should be positions(j.toInt)(row((i - j - 1).toInt).toInt)(row(i.toInt).toInt) += 1

@lyxal ^

lyxal

oh

oh ffs

can't believe I've done this

almost there

now there's just one last thing wrong with the bits generated :p

1 2 3 4 5 6 7 +++++ comes out at 123 bits normally, 135 bits with the scala port

but the bottom and top calculations are correct

AndrovT

what are the exact outputs?

lyxal

normal:

010010000111010010000011100111001001011101001101001010001110101111111001100110101101100100001000001111111110000100101101111

scala:

010010000111010010000011100111001001011101001101001010001110101111111001100110101101100100001000001111111110000100101101111000000000000

hopefully it's just type conversion problems

AndrovT

github.com/Vyxal/Vycoder-Scala/blob/… => for i <- 0 until (bits - (top - bottom + 1).bitLength) do out += 0

lyxal

I don't know what I'd do without you :p

thanks!

actually, thanks for all of this! I really appreciate the effort you've placed into vycoder and the patience you have with helping me make a version that can be used for the web :p

there we go

works a charm :p

time for me to write the js bridge part

luckily that's just copying the v3 js code and changing it a little :p

bridge is working good

I might start on the decoder then

lyxal

13:37

well the js works

it's a tad slow, but that's to be expected of js

at least the byte count can be determined online

even if the live count is a little slow

Tomorrow I will take the js generated by scala.js and connect it to the site (locally) as well as the python

"but lyxal how do you call js from python?"

there's a library for that

Apr 24 at 9:05, by lyxal

https://pypi.org/project/Js2Py

I'd like to thank everyone (Androv, user and steffan) for helping out with the process

y'all are the big brain geniuses that keep this project going

all of y'all are (not just those who've helped with vycoder)

Hey maybe the website can preprocess bit strings to save the server some time

So like if you press run, it automatically converts before sending a request

It'd strip the flags too

And the flags in the interpreter code would only run offline

naffetS

23:35

Should we propose adding Vyxal to Exercism?

We could just post on forum.exercism.org and if they agree then I suppose we can be track maintainers and work on itt

Someone sstarted on github.com/exercism/05ab1e but no one is doing anything rn

lyxal

I'm for it if it's allowed

naffetS

Anyone can propose a new track

exercism.org/docs/building/tracks/new/request-new

There's also a discord but the forum is better to discuss that type of thing

Transcript for