The Nineteenth Byte

Rust, score= ~6,081.9 Guesses each number 1-9. Aborts if the number found is less than 0.8 * number. use rand; use rand::{Rng, RngCore}; struct Which { rng: rand::rngs::ThreadRng, numbers: Vec<f32>, score: f32 } impl Which { fn new(rng: rand::rngs::ThreadRng) -> Which { ...

graffe

@thejonymyster that sounds much better

thejonymyster

yeah im glad it was just that and not something weirdly wrong about the rest of my code

mousetail

That's really close the the analytically computed score

Number Basher

2:22 PM

@mods maybe move these?

lyxal

Make a room and I will

graffe

@mousetail did you try optimizing over the constant 0.8?

Number Basher

@graffe you or me

graffe

that is trying different values

mousetail

@graffe Nope, just random guess

graffe

2:25 PM

@NumberBasher sorry?

mousetail

I'm sure changing it would improve the score

graffe

@mousetail ah.. well that's an easy improvement :)

mousetail

Also using a step size other than 1.0 would probably improve it

graffe

all sounds good to me

Number Basher

@graffe i mean, who make a new room for your challenge

graffe

2:27 PM

on discord?

lyxal

on SE chat

mousetail

No on here

graffe

could you please do it

I will have to go in a little bit to go to a birthday party!

but what is wrong with chatting here?

Number Basher

well this is tnb, not "graffe's challenge discussion"

thejonymyster

@graffe omg have fun yay

@NumberBasher tnb is about cgse, which contains graffe's challenge :P

Number Basher

2:29 PM

doesnt mean its the best place

graffe

@thejonymyster thanks!

thejonymyster

fair point :P idk then lol im just riffing

mousetail

Is the discussion whether this topic is off topic itself off topic?

3

caird coinheringaahing

I think the conversation is fine here

If it starts pushing out more relevant discussion, then sure, take it elsewhere, but it's fine atm

graffe

@cairdcoinheringaahing thanks

Number Basher

2:32 PM

oh ok

graffe

@mousetail maybe the various parameters can be plugged into scipy to be optimized?!

mousetail

@graffe Maybe

graffe

@mousetail :)

mousetail

You'd have to find a analytical way to compute the score though

I'll leave that to others

graffe

@mousetail can you just estimate it by running it a thousand times?

mousetail

2:34 PM

Yes, but not witth SciPi right?

You could do some kind of biased random walk algorithm

graffe

@mousetail why not? I mean define a function score that takes some parameters and returns the estimated score

then optimise that function over the choices of parameters

mousetail

Oh you mean fitting the expected score to a curve with some kind of non-linear regression then finding the minima

Maybe possible but seems very hard compared to biased random guessing

graffe

@mousetail I wasn't thinking over anything so clever. I literally meant have scipy optimize a function that you define that takes some parameters

there are methods that don't need derivatives

just anything function that returns a number will do

Number Basher

feedback welcome https://github.com/TvoozMagnificent/SunSip

graffe

just in our case the number will be an estimate

Number Basher

2:38 PM

bye guys

just ping me for anything

mousetail

@graffe How do you find the function though?

graffe

@mousetail as an example, the function takes in a single number and runs the simulation for 1000 iterations. Let's say the number is the value of 0.8 in your code so a real between 0 and 1

mousetail

You can't optimize a function like that in scipi I think

graffe

the simulation gives us the estimate of the cost for a particular value of that number

mousetail

It needs to be differentiable

graffe

2:40 PM

@mousetail It doesn't. They have fancy things now like shgo

docs.scipy.org/doc/scipy/reference/generated/…

mousetail

Ok that's probably the biased random walk I was talking about

graffe

ok :)

let's bet what the best score will be?

I am going to go for 2000

graffe

2:59 PM

hmm.. no bets?

mousetail

3:24 PM

@graffe I'll bet 1 (one) rep that I don't win

Steffan

3:58 PM

I guess I never noticed the meta post said the nominator is supposed to post the post

thejonymyster

my strategy of "do the naive strategy with a number less than 9" doesnt seem to be getting me anywhere

mousetail

4:26 PM

@thejonymyster Try a different number, but if you are close enough try a bigger number

graffe

4:39 PM

@mousetail but how do I give it to you when you do!

thejonymyster

@mousetail oh yea that makes sense lol

mousetail

5:15 PM

@graffe So a accept is 15 rep and a downvote 2 so all you need to do is accept my answer then downvote is 7 times.

Steffan

5:26 PM

or just upvote once, then you have to downvote 9 posts yourself.

(if you downvote an answer, you lose 1 rep)

mousetail

Yes that's probably easier

now how do I found 9 down vote worthy posts

caird coinheringaahing

5:43 PM

Or, just accept and downvote one answer

Feeds

5:57 PM

0

Q: Language of the Month for September 2022: Prolog

In accordance with our meta agreement, since one candidate received more votes than the others, we have a new featured language! Throughout September 2022, our Language of the Month will be: Prolog What's a Language of the Month? See the meta post for nominations. In short, during September, th...

mousetail

6:15 PM

@Feeds Lost too much rep last time that you are changing it from 3rd answer to 5th?

Steffan

6:32 PM

no it's just that prolog is easier to program in than knight

so it's easier to write more answers

(imo)

mousetail

You think? Maybe IDK

Steffan

I only lost a couple hundred, wasn't much

Knight is like a bare minimum language, you know, no builtins for anything

And no lists

Prolog is more normal, builtins for all sorts of stuff

IMO, much easier to write code

I guess the rep is part of it, because I'm so close to 10k

user

6:49 PM

I can start bounties if you like

Or instead I can award a couple +500 to strffan or someone in case I’m not around to award them

@Steffan how much do you think you’ll need? I can spare like 4000

Oh wait I can spare more than that, noice

user

7:06 PM

@hyper-neutrino @WheatWizard Could you unfreeze the Prolog room? Thanks and sorry for the pings

New Posts

0

Q: Detect round trips on a dodecahedron

An ant starts on an edge of a dodecahedron, facing parallel to it. At each step, it walks forward to the next vertex and turns either left or right to continue onto one of the other two edges that meet there. A sequence of left/right choices that returns the ant to its initial state (edge and dir...

code-golf math geometry graph-theory

Newly Featured Posts

9

Q: mini SPICE solver

SPICE is an electrical circuit simulation program originating from UC Berkeley. Your goal is to implement a minimalistic version which can solve for the nodal voltages and branch currents of a circuit composed of resistors and DC current sources. Input Format The input may come from a file, io ...

code-golf parsing electrical-engineering

36

Q: Sum of Modulo Sums

Given an integer n > 9, for each possible insertion between digits in that integer, insert an addition + and evaluate. Then, take the original number modulo those results. Output the sum total of these operations. An example with n = 47852: 47852 % (4785+2) = 4769 47852 % (478+52) = 152 47852 ...

code-golf math number-theory

forest

7:30 PM

Say, would anyone here happen to know the best way to optimize some trivial assembly?

	ldh [32]
	and #0xf3f
	jeq #0x01, drop
	jeq #0x02, drop
	jeq #0x04, drop
	jeq #0x08, drop
	jeq #0x10, drop
	jeq #0x11, drop
	jeq #0x12, drop
	jeq #0x14, drop
	jeq #0x18, drop
	jeq #0x19, drop
	ret #1
drop:
	ret #0

I can't think of any efficient way to optimize this since jumps are only allowed forward (language is cBPF).

graffe

@mousetail :) I will resist the temptation

user

@forest Depending on what you're optimizing for, you can post it as a tips question on the main site (disclaimer: may fuel the current controversy ;P)

forest

@user Size (actually performance, but size has a nasty overhead in this). These are TCP flag combinations.

user

That sounds like an objective winning criterion to me

forest

hm

user

7:35 PM

If you post it as a question there are definitely people on CGCC who know assembly who'll see it

forest

It's not CG though.

user

Here it's less likely to be solved

forest

Just optimization of a regular language.

Yeah I know. Was asking just in case anyone happened to have a quick idea.

user

@forest If you're shortening it it's just code golf right

forest

Perhaps. I'm not sure since I haven't posted to CGCC before. :P

user

7:38 PM

@forest What are all the possibilities? Maybe swapping #1 and #0 would let you have fewer jeqs?

forest

What do you mean?

All the possible instructions? Flag combinations?

user

Wait I misunderstood

emanresu A

@lyxal Probably :p

user

I thought you're checking if one single item on the top of the stack is 0x01 or 0x02 or ... (I was asking what possible values that could have)

emanresu A

Although I'm nomnating you

user

7:39 PM

I think you can only nominate yourself?

forest

@user It loads a half-word from address 32 into the A register, ANDs it with 0xf3f, then checks it against all those other immediates.

emanresu A

@user No?

user

Okay, so it is just one thing

@emanresuA Oh ok

emanresu A

I don't really know either but it seems sensible

user

@forest What are all the possible values for that half-word, I meant?

forest

7:41 PM

@user All possibilities within the limit of the bitmask 0xf3f.

emanresu A

@WheatWizard Would you mind putting prolog LoTM on the upcoming events board?

caird coinheringaahing

You can always raise a flag

emanresu A

True

Sandbox Posts

9:16 PM

0

A: Sandbox for Proposed Challenges

Persistence of a number Given a blackbox function \$f : \mathbb N \times \mathbb N \to \mathbb N\$ and a positive integer \$x = d_1d_2d_3...d_n\$, where \$d_i\$ are digits with \$d_1 \ne 0\$, output the persistance of \$x\$ under \$f\$. Shortest code wins

Steffan

9:46 PM

@user keep your rep lol

caird coinheringaahing

@thejonymyster I appreciate the feedback, but I've posted ^^ to get it down on paper, not to get feedback :P

It's a lot easier to edit an existing post than it is to have a draft to work on

thejonymyster

Well i guess ill explode then

nah i get u ^_^)b

excited to see it fleshed out then, take ur time

Steffan

10:01 PM

typo: persistance

user

10:28 PM

@Steffan I don't want or need it

Bubbler

11:22 PM

@forest seems the logic can be factored into something like this (depends on instruction support though):

if x == 0x10 or 0x19: return 0
if (x & 0xf20) != 0: return 1
x = x & 0xf
if x == 0x1 or 0x2 or 0x4 or 0x8: return 0
return 1

forest

@Bubbler Hm, OK I'll try writing it like that and see if it's shorter.

The first "if x == 0x10 or 0x19" probably wouldn't work because the and #0xf3f is used to mask out bits that should be ignored, so I'd have to still keep that AND in there.

Bubbler

Oh, I meant after x = x & 0xf3f

forest

Since there's only one general-purpose register and one index register, doing if (x & 0xf20) might require an extra instruction or two.

Er, actually no, I can use jset for that.

Yeah it looks like that saved two instructions. I'll test it out.

(It'd be easier if it had some sort of popcnt, but it doesn't)

	ldh [32]
	and #0xf3f
	jeq #0x10, drop
	jeq #0x19, drop
	jset #0xf20, match
	and #0xf
	jeq #0x1, drop
	jeq #0x2, drop
	jeq #0x4, drop
	jeq #0x8, drop
match:
	ret #1
drop:
	ret #0

Bubbler

11:41 PM

If the machine was something closer to x86, I could suggest something like selecting x-th bit from a 32-bit constant (the ten positions are 0, the rest are 1)

Transcript for