The Nineteenth Byte

Most likely. I use the same method to score and draw right now, so it makes it much faster to not do that while scoring. Might separate them out and retouch it a bit after finding points.

Eric Tressler

it's still cool. I'll read your placement algorithm when you make your post

since that's the only really interesting part of the problem, to me

Geobits

I wish I could say it was a more interesting algorithm. It's pretty naive, just uses a lot of iterations.

Martin Büttner

@EricTressler it is supposed to be the core of the challenge after all :P

Geobits

Seems to work all right, though.

I thought the core was posting pretty pictures in chat. Seriously, I think I've been doing this wrong.

Martin Büttner

12:37 AM

:D

Martin Büttner

12:48 AM

oh, we've got 3,333 questions

Eric Tressler

Well

when you consider that we have 110100000101_2 questions, that seems less impressive

PhiNotPi

1:00 AM

My sister got some new goldfish today. They are named Kaede and Wilmington.

Eric Tressler

is that a reference to something

PhiNotPi

I honestly do not know.

orlp

1053 points

that's starting to look good :P

@MartinBüttner ^

PhiNotPi

Are they non-random points?

orlp

it involves random sampling

but with heuristics

Eric Tressler

1:21 AM

haxiomic.github.io/GPU-Fluid-Experiments/html5/?q=Medium

I want to make something like that, but I can't even figure out the basic setup that's using. I want to make a local copy of it so i can rip the guts out and make my own variant

because i think that is a pretty cool demo.

but i go to the source on github, and it's a bunch of .hx files

and java. :/

orlp

1:35 AM

1453 cells

Martin Büttner

@EricTressler Ah yeah that's an amazing simulation... I'd actually like to do a PhD in fluid animation

orlp

@MartinBüttner they're starting to look real smooth :)

Eric Tressler

I've been trying to find javascript GPU particle examples, but for some reason none of them are running properly here

orlp

the longest time spent is actually computing average colors =/

Eric Tressler

but I looked at some of the math behind that demo, and it's pretty complex. I think it does a great job with turbulence and viscosity, though

Geobits

1:51 AM

@orlp Devon approves of your version of his portrait :P

Alex A.

2:17 AM

It's so cool seeing people's Voronoi versions of my girlfriend's painting! :D

Sp3000

Oh boy, I have a lot of catching up to do :P

Alex A.

2:31 AM

@Geobits: Your son is famous!

Calvin's Hobbies

2:44 AM

My river photo is excluded :(

Alex A.

@Calvin'sHobbies: Doesn't mean you can't do it. If I recall, the challenge says you can try it on others.

Calvin's Hobbies

@AlexA. I just meant that it isn't one of the defaults even though it has been in a number of image challenges. I'm not that bent out of shape about it :P

@Doorknob (and everyone else) Has there ever been discussion about adding new posts to the main page to the chat feeds?

I mean like how meta and sandbox posts are autofed to chat

We don't get that many questions and it seems useful.

Alex A.

I second this ^

Calvin's Hobbies

@MartinBüttner What does the other mod think?

Alex A.

3:01 AM

@Calvin'sHobbies: Your honeycomb challenge has been up less than 30 minutes and it already has two answers. :)

Calvin's Hobbies

@AlexA. Good gracious!

orlp

952 cells

Alex A.

@orlp That's legit! Nice work!

Calvin's Hobbies

3:25 AM

@AlexA. On second thought the feeds take forever...

Alex A.

What do you mean?

Calvin's Hobbies

My question hasn't been auto posted yet

Alex A.

Oh yeah. How long do they typically take?

orlp

is there a difference between the geometric average of L*a*b* color coords and the 1-mean clustering of these coords?

Calvin's Hobbies

Not sure. I've been looking for a mother meta post that mentions it

(That's not for you orlp :P)

Alex A.

3:28 AM

Haha

orlp

hrm

I need a KDTree that can handle updating data

Alex A.

@Calvin'sHobbies What's a "mother meta post"?

Sp3000

I need a range tree but scipy.spatial only has KD :(

Calvin's Hobbies

@AlexA. From meta, not meta

Alex A.

Oh! I always forget about that one.

orlp

3:31 AM

1600 cells

Alex A.

@Sp3000 Let me play you a song on the world's smallest violin.

New Meta Posts

2

Q: Do we want new challenges auto posted in chat?

Our chatroom The Nineteenth Byte currently has 3 message feeds. For example, when there is a new meta post, it is automatically posted in chat. Do we want to have new questions from the main page be a new feed?

Alex A.

@Calvin'sHobbies: Guess we have our answer. About 20 minutes.

orlp

hmm

Alex A.

@Sp3000: Why don't you write your own?

orlp

3:32 AM

the escher one is really hard to express nicely

Alex A.

Your hedgehog looks great, @orlp.

orlp

fun fact, if you use 8 bit color that's a ~4.8kb compressed image of the hedgehog :P

Alex A.

:O

orlp

1600 points * 2 bytes for position * 1 byte for color

Eric Tressler

4:54 AM

anyone around?

orlp

yes

Eric Tressler

does soulwire.co.uk/experiments/webgl-gpu-particles work for you

I get a single white point on my screen, nothing moves

and i have a pretty decent graphics card. I can't figure out why these javascript particle demos aren't working; i tried firefox and chrome, both

xnor

i see same as you

orlp

for me it 'works'

Eric Tressler

okay. well, thanks for checking. it's supposed to be a bunch of particles doing something, but... who knows

orlp

5:04 AM

as in the cursor spawns dots

but they dont move or something

Eric Tressler

oh. that happened to me with chrome, orlp. I think maybe the demo is either broken or making some weird assumptions

i'm trying to learn to use the GPU in webgl, because I have some graphics-intensive demos that I think could be like 100-1000x faster if I would just learn this

orlp

3500 cells

(with markers to show how good the sampling is :P)

Eric Tressler

looks pretty good

not very ferocious-looking

Calvin's Hobbies

5:37 AM

I like how the bear looks even fuzzier

Eric Tressler

in the "thumbnail" it looks like a stuffed teddy bear

New Sandboxed Posts

6:48 AM

0

A: Sandbox for Proposed Challenges

(Currently unnamed) code-golf Introduction Arithmetic Gaol is a special facility that incarcerates positive integers. However, recently, the positive integers have been trying to escape. Therefore the wardens have decided to eliminate some of the positive integers to send a message to the othe...

Calvin's Hobbies

7:35 AM

Say that 5 times fast

Sp3000

My money's on another mod-chain being the winner :D

xnor

the "any other string is 0" requirement makes this tougher than previous recognition challenges

Optimizer

till then, I am posting the most trivial solution

Sp3000

Oh anything else needs to give 0. Damn :P

xnor

basically, any code has to contain as much information as the list of strings, since they are determined by the code

this might turn into a yawn compression challenge

Calvin's Hobbies

7:38 AM

Hope not :/

xnor

@Calvin'sHobbies I'm not sure this counts as kolmogorov-complexity -- it takes input

Calvin's Hobbies

True but you have to encode the words

Sp3000

btw @xnor very nice honeycomb

xnor

thanks!

Sp3000

I was wondering if it was possible to shorten max(7-i,i-24,0) with some sort of insane bitwise method, but I'm not sure how to easily find such a solution

xnor

7:41 AM

there's so many degrees of freedom available here

i can be shifted, and everything can be shifted by a constant

Calvin's Hobbies

Would you guys prefer if it was If any other string is input, anythings besides 1-9 should be output? (could be non-numeric, can be diff across inputs)

Still not terribly better perhaps

xnor

either way, I think answers will need to show that no other input can create a collision

Sp3000

Does that make a difference? The only thing I can think of is that some languages give -1 for .find()

Calvin's Hobbies

You're right, doesn't really help.

I think I'll just remove the 0 thing. Hang on

xnor

ok, mod chains go go go!

Calvin's Hobbies

7:57 AM

Ok

Optimizer

are you talking about mod chaining for the 9 similar words challenge ?

xnor

yes

Optimizer

its going to be really hard to check that any other possible word does not give the same result

xnor

CH changed it to only caring about the 9 given words

it can do anything on other strings

Optimizer

ohhh

Sp3000

8:04 AM

Mod chain gogogo indeed :D

xnor

maybe there's a golf or fastest alg challenge to be made about mod chains

Sp3000

Input: max base, max mod, max number of mods

Optimizer

11 bytes

still golfing

Jakube

8 bytes

xnor

wow

sum(map(ord,input()))%x%10?

Optimizer

8:10 AM

11

Jakube

Yes, only difference is that in Pyth I can do ord(input()) directly.

xnor

how do you do that?

Optimizer

what is x here ?

xnor

some number to be found by brute force search that makes the outputs distinct

Optimizer

haha lol

so 11 works

except it gives 11 for something instead of 9, all other are fine

orlp

8:13 AM

I'm almost done with my image code

just putting the final touches here and there and doing the rendering

Sp3000

Must... be.. as.. good... as... orlp

orlp

I guess I can let the cat out of the bag

Optimizer

@Jakube you are modding with 8109 ?

Jakube

Yes

Optimizer

what is e%

orlp

8:15 AM

I'm using Poisson weighted disc sampling, and averaging colors in L*a*b* space.

the weights are chosen by a combination of two heuristics: edge detection and local entropy

Jakube

two different commands. % 8109 does mod 8109, e gives the last digit (mod 10)

xnor

didn't know that about e, that's perfect here

how does C work a string?

does it treat it as base 256?

orlp

it's Python's ord

Optimizer

weird. I get different result

Sp3000

@orlp Are you doing anything fancy for edge detection? I'm lazy so I'm using the poor man's edge detect over here :P

xnor

8:18 AM

@orlp but Python ord doesn't work on strings, right?

orlp

@xnor it only works on strings

@Sp3000 I'm using Sobel edge detection

xnor

ord('abc') gives me an error?

orlp

@xnor it only works on single characters

Sp3000

Ah, that makes sense :) maybe if I get time...

orlp

oh wait

C

Jakube

8:20 AM

@xnor It does map(ord, input()) and interprets the list as base 256.

xnor

ok, makes sense

orlp

@Sp3000 and entropy + edge detection gives you this sort of sampling density:

Sp3000

orlp

needs some work :P

althought mine is more than 1000 cells

Sp3000

Hm... I think I need to work on the edges

orlp

8:23 AM

on toilet right now, can like a 1000 cell one in a bit to compare

Sp3000

Yeah I still need to do some pruning :P Mine still puts a bunch in the black space of the planet/nebula pics currently

Hmm yours looks nice and even with the close points

orlp

@Sp3000 poisson disk sampling

already told you :P

Sp3000

It doesn't mean anything to me if I don't know what it is :P (I'll look it up later)

orlp

@Sp3000 how did you render all your images? by hand?

Sp3000

With the snippet

orlp

8:31 AM

yes, but I mean by hand copy pasting

Sp3000

Oh right, yes

orlp

as promised, the 1000 cell bear

Sp3000

At least you have an eye :P

orlp

and ears :P

Calvin's Hobbies

@orlp Very nice. I'm eager to see your answer. What's the language?

orlp

8:35 AM

@Calvin'sHobbies python

Calvin's Hobbies

I'd like to see the bear at 100 to compare to martins

orlp

one sec

@Calvin'sHobbies with or without dots?

Calvin's Hobbies

without I guess

But it's not a big deal, I can wait for the answer

orlp

it's not great

but I'd say better than martins

Sp3000

It's easy to be better than Martin's because the color's easy to mess up :P

VisualMelon

8:39 AM

colours are certainly more representative ;)

orlp

people need to learn you can't average RGB values :P

Sp3000

Martin doesn't even average, he just takes the colour

orlp

ah

Calvin's Hobbies

At what N could a stranger to the contest tell it was a bear :P

Sp3000

300 maybe?

orlp

8:53 AM

5000 cell hedgehog

Sp3000

Oi 5000 is cheating :P

orlp

show me your 5000 hedgehog :P

Sp3000

k, I'll try running it while I do some housework (was in the middle of tweaking parameters so I hope this doesn't turn out too bad :P)

MemoryError <--- :/ I need to clean up my disk space

orlp

memory errors? :P

what are you doing

Calvin's Hobbies

What if this is input with N = 20?

Optimizer

9:01 AM

lol

Sp3000

More like why do I only have 1GB of disk space left

Optimizer

pom

orlp

sec, coming in

not too great :P

at best it keeps the colors approximately the same

Sp3000

@Calvin'sHobbies Will try in a bit but I'm not sure if the black dots might interfere. Will see :)

orlp

n = 500

5000 cells

Calvin's Hobbies

9:09 AM

It turned into a person!

Sp3000

Close enough :P I just figured out why I have a bias against the left edge though, fixing.

orlp

funny, yours is much much better at this one

Sp3000

Even funnier - I just realised I had my sampling backwards :P

Oops

orlp

oops, my yoshi is messed up

Sp3000

Wait nvm it was right

How's yours for Escher? I'm still finding that the hardest

Calvin's Hobbies

9:18 AM

Ooh, how about this, N = 7 (you guys can stop if you're sick of posting images)

Sp3000

Nah this is good fun :D

Unfortunately mine take a lot longer to run than orlp's though

Calvin's Hobbies

Actually the concave shapes won't ever work will they

orlp

mine is <1 sec for small sizes, <5 sec for 5000 cells

Sp3000

Mine is dependent on number of cells, but more importantly the number of sample pixels to k-means

^^ n = 500 is done

orlp

also better than mine

mine doesn't like straight edges it seems

Sp3000

9:21 AM

I think mine are better at paintings/flat colours than photos

orlp

and that's a direct consequence of circle sampling

Sp3000

Yeah this is pretty impossible

Calvin's Hobbies

How about this (N around 32?)

Or N much larger

Sp3000

(gimme a sec, running something else atm - having only 5 colours might make this interesting though)

32's a bit too low I think, the colours get really messed up

Oh I know what I did wrong now - random.triangular defaults to having the mode at the midpoint, so I've been sampling from points which are neither extreme edge points nor points which are very similar to surrounding colours locally

Calvin's Hobbies

10:04 AM

@Sp3000 still pending?

Sp3000

Doing a bit of tweaking and testing. Basically N = 100 was practically unrecognisable, and 500 was still quite messy

Mine's not great when you have similar colours in different locations, unfortunately :(

Calvin's Hobbies

Haha, yikes :P

So much black on the edges but not on the other borders

Sp3000

That's because the centroids are naturally closer to the centre of the image, so any colours on the outskirts tend to be a bit larger

orlp

holy finally

figured out this stupid bug =/

Sp3000

Oh?

orlp

10:22 AM

Sp3000

5k is still cheating :D

Calvin's Hobbies

But where's the black background?

Sp3000

It's the "challenge" version below the pics

Yoshi was originally on a white background, but was changed because it makes the belly significantly harder

Calvin's Hobbies

ah ok

Sp3000

10:45 AM

@orlp Still haven't done the getting-rid-of-useless-cells thing yet and I had to drop the sample points because of MemoryErrors, but anyway for comparison ^^

Martin Büttner

11:06 AM

@sirpercival your results look quite nice :) ... you'd probably get some more attention if you included an explanation of the algorithm and more thumbnails in the post itself.

Optimizer

11:17 AM

he's not doing it for the fame or attention

Martin Büttner

I'd still be interested how it works :P

Optimizer

oh you paparazzis

trichoplax

In a popularity contest I'm at least as swayed by an interesting explanation as I am by the output

Optimizer

^

trichoplax

(I agree that plenty of output thumbnails that don't require going off page is important too)

Especially once there are dozens of answers to scroll through...

Sp3000

11:22 AM

btw should we also give lots of the text output so people can play around with it?

(I haven't done that yet because I'm still fiddling)

trichoplax

Even just one or two text output examples will make for a lot of scrolling...

Sp3000

I'm pastebinning mine :P

trichoplax

Yeah that's one thing I would prefer to see linked off site

sirpercival

11:40 AM

good morning, all

@MartinBüttner i didn't do either of those things because I was headed directly bed afterwards. :D I'll put more stuff in right now

Martin Büttner

no worries :)

sirpercival

should i include all 36 thumbnails i have (not including the original images)?

or is that overkill?

Martin Büttner

if you use the small ones, I don't see a problem

sirpercival

ok then, that's what i'll do

Martin Büttner

11:53 AM

added a voronoi diagram to the test gallery

@orlp can you actually specify the number of cells now?

(also yay for Poisson disc sampling... that's still on the list for Random Golf of the Day)

Sp3000

The voronoi has a bit of a white border on the right and bottom

orlp

@MartinBüttner yes

@MartinBüttner I did some funky math

Martin Büttner

nice :)

@Sp3000 oh crap... wait

orlp

5000 cells, but def. my favourite :P

sirpercival

oh i forgot to do the white yoshi

orlp

11:55 AM

@MartinBüttner is there an easier way to render everything than to copy paste and screenshot all the time? =/

Martin Büttner

I don't know... writing your own renderer? :/

sirpercival

@orlp use IDL with its native graphics output :D

Martin Büttner

can you save the svg and then rasterise it locally?

orlp

@MartinBüttner I can't right click to save

it's not an image

Martin Büttner

oh okay... I don't know how browsers handle embedded svg

orlp

11:58 AM

@MartinBüttner basically I got the math right that if I linearly increase my 'parameter' it linearly increases the number of cells made (within a few percent)

so I just do a 'calibration' run before generating images :P

Transcript for