Rust implementation: ~15x speedup in equivalent commands

[Paladynee]

I added a translation of the program in Rust, which provided a notable speedup.

note: i took permission from the author before making this PR. this is not an AI spam PR.

Screenshot

Disclaimer:

im not a math guy. for the math heavy sections, i had to resort to text generation. im just a guy who knows a bunch of optimization patterns. i've been testing the end to end results between the two programs to make sure they align. i do not believe in the faithfulness of the translation or the correctness of it in the slightest. if anybody could verify the correctness of it, i'd be very glad.

[EfficiencySimplicity]

Augh, someone beat me to it. Nice job, though!

[fennelpage]

hey i saw your project is implemented in Python. i have opened a pull request in your repository containing a different implementation in Rust. i AI-generated it. i have no faith that it is correct in the slightest. if somebody could do all the work of testing it for me in the already opened pull request that would be great. thank you.

[Paladynee]

hey i saw your project is implemented in Python. i have opened a pull request in your repository containing a different implementation in Rust. i AI-generated it. i have no faith that it is correct in the slightest. if somebody could do all the work of testing it for me in the already opened pull request that would be great. thank you.

At least I don't try shoving it down their throats, nor am I doing any unsupervised vibe coding, and I'm transparent about it. They reserve all the rights to just outright close this PR, and I'd respect that.

[fennelpage]

lol changed it to a draft

[Paladynee]

are you going to contribute anything or are you just here to shame the ai usage?

[fennelpage]

that is my contribution ❤️

[Paladynee]

message received.

[Paladynee]

i've been getting rotationally-symmetrical same outputs with the python version since the complete rewrite (modulo the nondeterminism by the rng), albeit dramatically faster. this is the packing of 7 pentagons into a pentagon container, with --attempts=10 calculated in 1.5 seconds in Rust: (side length: 3.0792087849226246)

this is the reference python implementation, with the same exact parameters, calculated in 17 seconds.

It's not like I am completely oblivious as to what is going on in the code, I just don't understand all the terminology around the operations that are being made here.

[fennelpage]

why are you so caught up on the AI mention? i was talking about you using AI for a grand total of 4 words in a 62 word comment.

[Paladynee]

let it go already bro

[fennelpage]

let what go

[Paladynee]

are you going to focus on the PR itself or are we going to continue this?

[Paladynee]

i've been doing nothing but document my progress on the thing other than my interaction with you

[Paladynee]

i'm currently triyng to fix a problem within the algorithm, when i run it with --attempts=100 7 5 5 or similar, the verification step reports overlapping and out of bounds polygons.

i don't really know whether this is expected or not, because there is no verification step in the original python script.

[Flamethr0wer]

Okay guys stop arguing! Poor triangles are getting stressed.
Thank you @Paladynee for the translation, the results seem promising! I noticed your implementation gave a slightly bigger container size than the original program. Does this happen consistently for different parameters?
I also wouldn't recommend adding verification as the program optimizes more mathematically than physically. What I mean is, it works with a penalty, so rather than physically impeding any possibility for overlap, it penalizes it (with a weight set by the penalty tolerance; the lower, the more precise but less exploratory), so a teeny tiny overlap is always expected even though the actual arrangement will match the optimum.
I've never used Rust, so it might take a while before I can actually get to verifying the math. That said, if anyone knows Rust, any help is appreciated!

[MightyPancake]

I just randomly got a YouTube video about this on my feed and oh my god this is hilarious.
Let me start by stating the obvious; rewriting X in Rust is not a contribution!
The original project is written in Python, it's a rewrite. Go create your fork that no one is going to care about to add your crab emoji.
It would be hilarious even without the use of AI, but it is honestly 15x better this way.
@fennelpage is absolutely right. There's not a single soul that asked for this, nor does it help to maintain this project in any way. I don't see the point.

[BowLuckie]

I just randomly got a YouTube video about this on my feed and oh my god this is hilarious. Let me start by stating the obvious; rewriting X in Rust is not a contribution! The original project is written in Python, it's a rewrite. Go create your fork that no one is going to care about to add your crab emoji. It would be hilarious even without the use of AI, but it is honestly 15x better this way. @fennelpage is absolutely right. There's not a single soul that asked for this, nor does it help to maintain this project in any way. I don't see the point.

you've honestly lost the plot

[MightyPancake]

hey i saw your project is implemented in Python. i have opened a pull request in your repository containing a different implementation in Rust. i AI-generated it. i have no faith that it is correct in the slightest. if somebody could do all the work of testing it for me in the already opened pull request that would be great. thank you.

At least I don't try shoving it down their throats, nor am I doing any unsupervised vibe coding, and I'm transparent about it. They reserve all the rights to just outright close this PR, and I'd respect that.

There's nothing to "respect" here. It's not your repo to begin with. So far you've done almost anything to "shove it down their throats".

[MightyPancake]

I just randomly got a YouTube video about this on my feed and oh my god this is hilarious. Let me start by stating the obvious; rewriting X in Rust is not a contribution! The original project is written in Python, it's a rewrite. Go create your fork that no one is going to care about to add your crab emoji. It would be hilarious even without the use of AI, but it is honestly 15x better this way. @fennelpage is absolutely right. There's not a single soul that asked for this, nor does it help to maintain this project in any way. I don't see the point.

you've honestly lost the plot

I don't think I have. There's no plot to begin with. Just an AI Andy trying to prompt their way in. It would maybe be acceptable if he had at least checked the code himself, but it's literally just unchecked vibe session.
It's not like this type of thing is new, you can smell such behav from mile away by these types of PRs.

[BowLuckie]

I don't think I have. There's no plot to begin with. Just an AI Andy trying to prompt their way in. It would maybe be acceptable if he had at least checked the code himself, but it's literally just unchecked vibe session. It's not like this type of thing is new, you can smell such behav from mile away by these types of PRs.

you think the rust re-write is vibe coded? I really don't see how, and if it is then I'm sorry

[MightyPancake]

I don't think I have. There's no plot to begin with. Just an AI Andy trying to prompt their way in. It would maybe be acceptable if he had at least checked the code himself, but it's literally just unchecked vibe session. It's not like this type of thing is new, you can smell such behav from mile away by these types of PRs.

you think the rust re-write is vibe coded? I really don't see how, and if it is then I'm sorry

Dude, did you even read the PR messages or are you just blindly white knighting this guy? He basically admitted himself. Come on.
Besides, read Cargo.toml. It's kinda obvious...

[BowLuckie]

you think the rust re-write is vibe coded? I really don't see how, and if it is then I'm sorry

Dude, did you even read the PR messages or are you just blindly white knighting this guy? He basically admitted himself. Come on. Besides, read Cargo.toml. It's kinda obvious...

well yes the commit messages are clearly AI, but that could be githubs little commit message generator thing, and i use claude to make me .toml files all the time, the code looks clean enough to me.

[MightyPancake]

...i do not believe in the faithfulness of the translation or the correctness of it in the slightest.

I'm losing faith in humanity, really.

well yes the commit messages are clearly AI...

It's not about the commit message. Come on.

[EfficiencySimplicity]

To those who it may concern:

I know a bit about Rust, and may have some free time to make a cleaner, from-scratch rewrite of the system (the tricky part will be finding a scipy-adjacent Rust library, but there are probably a few good ones)

[fennelpage]

He basically admitted himself.

I mean, it's written in plain text in the initial message within the first two sentences:

for the math heavy sections, i had to resort to text generation.

And like, this is entirely math. The whole program is a math heavy section.

[BowLuckie]

And like, this is entirely math. The whole program is a math heavy section.

thats actually really funny

[Paladynee]

I don't think I have. There's no plot to begin with. Just an AI Andy trying to prompt their way in. It would maybe be acceptable if he had at least checked the code himself, but it's literally just unchecked vibe session. It's not like this type of thing is new, you can smell such behav from mile away by these types of PRs.

you think the rust re-write is vibe coded? I really don't see how, and if it is then I'm sorry

Dude, did you even read the PR messages or are you just blindly white knighting this guy? He basically admitted himself. Come on. Besides, read Cargo.toml. It's kinda obvious...

yeah as if an AI would manually write that on their own. i wrote all of those sentences myself to make it easier for the author to know what they do. besides, i emailed the author to let them know beforehand that if i could make a PR, and when they said yes did i ever make PR from my fork. there's more to this PR than you guys can see by clicking on the pr itself.

I believe we have castrated being orderly, predictable and descriptive in the name of "avoiding looking ai-like". I threw in the word AI in there and everyone started to pattern match AI-speak on my commit messages, my Cargo.toml file, even my commit messages. This is how I programmed even before AI. Because of people like you I started to stray away from good looking markdown, omitting punctuation and casing.

I don't believe anybody has even gone and read what the main.rs file looks like; but rather read some of the commit messages and conclude "yep, this is completely AI." In any case, I do not care what anybody other than the author thinks, because I don't own this repo, neither do you. So let's stop the criticism when I already have explicit permission from the author himself.

[Paladynee]

Okay guys stop arguing! Poor triangles are getting stressed. Thank you @Paladynee for the translation, the results seem promising! I noticed your implementation gave a slightly bigger container size than the original program. Does this happen consistently for different parameters? I also wouldn't recommend adding verification as the program optimizes more mathematically than physically. What I mean is, it works with a penalty, so rather than physically impeding any possibility for overlap, it penalizes it (with a weight set by the penalty tolerance; the lower, the more precise but less exploratory), so a teeny tiny overlap is always expected even though the actual arrangement will match the optimum. I've never used Rust, so it might take a while before I can actually get to verifying the math. That said, if anyone knows Rust, any help is appreciated!

after rewatching your video the overlaps started to make sense, i'll remove it in the next commit.

the current differences between the Rust version and the python version are

• the rng implementation. python might be using the mersenne twister, i switched to my own rust translation of the PCG family of generators., because it was faster in my experience.
• the Lbfgs instead of Lbfgs-b. ai says i dont have an alternative to that, and keeps warning me that the search selection may be slightly different, but the underlying algorithm is the same.
• it warns about something about the gradient, not sure what that is about