The Nineteenth Byte

flawr

4:00 PM

AAahrg, I know I'm gonna stumble over this =< a thousand more times.

Heureka!

Fatalize

rule: the = touches the pointy part :p

flawr

> touches the pointy part

=)

quicksort([],[]).
quicksort([P|L],S) :-  partition(P,L,A,B), quicksort(A,X), quicksort(B,Y), append(X,[P|Y],S).

^ this works!

TimmyD

> touches the pointy part

( ͡° ͜ʖ ͡°)

flawr

@TimmyD On a scale from 1 to 10, how old are you???

Fatalize

@flawr And it's readable yet short drool

flawr

4:05 PM

@Fatalize I tried something else, that always produces out of stack errors, do you see why?

quicksort([],[]).
quicksort(L,S) :- [P|_] = L, partition(P,L,A,B), quicksort(A,X), quicksort(B,Y), append(X,Y,S).

OOoooh I just found out.

Fatalize

lol

aint event got time to read the code he found out

TimmyD

@flawr Like, is this a logarithmic scale?

flawr

@TimmyD Nope, exponential

TimmyD

Uh ... what?

flawr

@Fatalize For me that's always a sign of progress=)

@TimmyD joking (unfunnily)

Bálint

4:09 PM

How would you guys define JavaScript?

I can't say it's an OOP language.

It's not really a functional language either

It's also not purely a prototype-based language, as you can create new objects

TimmyD

@flawr If it's powers-of-two based, I'm slightly over 5

Fatalize

crap (noun)
2
sometimes vulgar : nonsense, rubbish;

Nathan Merrill

@Bálint imperative

flawr

@Fatalize I was just trying to think of something like that=)

Bálint

@Fatalize ಠ್ಗಠ

Nathan Merrill

4:15 PM

@Bálint a language fits into lots of paradigms, and most languages rarely perfectly match a given paradigm. That said, a given paradigm more often depends on the programmer than the language

Bálint

@NathanMerrill Yes, but wich categories does it "contain"

I know it can be functional, because....functions...

Nathan Merrill

nearly all of them listed here:

Programming paradigm

In computer programming, the notion of programming paradigms can have somewhat different meanings depending on its context of use. Involving programming languages and programming: it is a way to classify programming languages according to the style of computer programming and idioms used. Involving software development: it is a fundamental style of programming, which is not generally dictated by the project management methodology (such as waterfall or agile). Paradigms differ in the concepts and abstractions used to represent the elements of a program (such as objects, functions, variables,...

Fatalize

@flawr your predicate crashes for a list of 10,000 zeroes

length(L,10000), maplist(=(0),L), quicksort(L,Z).
ERROR: Out of local stack

not sure there's an easy fix for it though

Bálint

@Upgoat I like the new header for your blog, but it bothers me, that it's very assymetrical

flawr

@Fatalize If you really want to sort 10000 zeros, you should probably use another language=)

(or realize that they are already sorted xD)

Fatalize

4:19 PM

or use a sorting alg that doesn't recurse that much

flawr

> or use a snorting algorithm that doesn't curse that much

TimmyD

So ... not Scarface

Fatalize

lol

I just used the msort builtin of Prolog to compare

it did it for 10000 immediatly

1 second for 1000000

not sure which algorithm it uses though

flawr

Well implementing quicksort that way in prolog is probably close to the most innefficient sorting techniques ever.

Fatalize

"The implementation is in C, using natural merge sort" well that explains it

flawr

4:24 PM

I bet they didn't use natural merge sort.

They sure used artificial merge sort.

Fatalize

^ weak af joke

throws tomato

Lembik

weak af?

TimmyD

GMO merge sort

Fatalize

@Lembik af = as bip

flawr

I recently thought about making a neuronal-net sort.

Lembik

4:28 PM

@Fatalize ah thanks

Nathan Merrill

actually, a neural net for a sort isn't too bad of an idea

I mean, lets say you are able to identify where you are being called from

Lembik

@NathanMerrill ? :)

Nathan Merrill

(your stack trace)

Lembik

hang on.. first you need as many outputs as inputs

Fatalize

@NathanMerrill No I'm Fatalize

damn you fixed it

mah joke

Lembik

4:31 PM

second a 5% error isn't going to work so well :)

flawr

@Lembik make it recursive then

Lembik

@flawr I think neural network factorizing would be more fun :)

Nathan Merrill

nah, you wouldn't use your neural net to match each element

Lembik

or primality testing

that I would like to see

flawr

@Lembik that would be interesting

Lembik

4:32 PM

I already know how to sort numbers :)

@flawr we need someone with a fast cpu or a gpu

flawr

I think the problem with all those is that neuronal nets usually only accept input of a certain size

Lembik

@flawr that's ok.. we can say it will always have 100 bits

or 10 :)

it's still very hard

Nathan Merrill

oooh, here we go. Your input is the current position (decimal 0 to 1)

flawr

@Lembik That is cheating then

Nathan Merrill

and the output is the predicted location

Lembik

4:33 PM

@flawr why?

standard crypto systems using a fixed number of bits

I still bet we can't do it :)

Nathan Merrill

you sort it according to the neural net

then do a secondary sort that sorts faster if the array is mostly sorted

Cᴏɴᴏʀ O'Bʀɪᴇɴ

@EᴀsᴛᴇʀʟʏIʀᴋ hahaha

flawr

@Lembik Then it is trivial

Eᴀsᴛᴇʀʟʏ Iʀᴋ

@CᴏɴᴏʀO'Bʀɪᴇɴ ikr

Lembik

@flawr what do you mean?

@flawr it's impossibly hard

so we are at different ends :)

Eᴀsᴛᴇʀʟʏ Iʀᴋ

4:35 PM

@avocadjuic 10/10 can i haz some juic

flawr

wel you only have a finite number of different inputs, so you're guaranteed to find a NN that produces the correct output.

Lembik

@flawr not if you are referring to a NN you can actually implement

If I have 2^100 different inputs you can't train a NN that recognises those in practice

flawr

@Lembik make an input layer with one node per bit

Nathan Merrill

if you have an array the size of 2^100, you have much bigger problems

Lembik

@NathanMerrill no.. 100 bits gives 2^100 possible inputs

Nathan Merrill

4:37 PM

ah ok

Fatalize

There's no point testing a NN on data you trained it with anyway, it will just learn the numbers directly instead of patterns

Lembik

@Fatalize my point is that you can't possibly train it with 2^100 differnet inputs

flawr

@Fatalize yeah that is what I thought too

Nathan Merrill

the guess here is that arrays to be sorted are often in similar patterns.

Lembik

we need a NN to play with

anyone got any code?

Nathan Merrill

4:38 PM

when called from the same location

flawr

So back to the sorting=)

Eᴀsᴛᴇʀʟʏ Iʀᴋ

@DrGreenEggsandIronMan your name runs of the edge of your flair. :P

Lembik

@flawr my idea is much more interesting :)

Nathan Merrill

therefore, you have a 1-input NN

where the input is the position in the array

Fatalize

@Lembik My point is there's no point to even train it with 10000 prime numbers, there are no patterns so if you want good recognition on the training set it will learn the numbers directly

Nathan Merrill

4:39 PM

and the output is the is the estimated final position

Fatalize

and won't generalize at all

Lembik

@Fatalize oh well that's a claim but do you know it is true? I mean do you know that there is no pattern?

@Fatalize I mean the pattern is that the numbers have no prime factors :)

flawr

@Lembik go train your nn and come back when you've found the pattern^^

Fatalize

@Lembik if you find a pattern that predicts primes numbers with 100% accuracy why are you wasting time here :p

Lembik

@flawr that was my point!

I would like to know if a NN can give a better than random output

Nathan Merrill

4:40 PM

the claim is that a sort call (in a real world situation) often passes arrays in a similar order

Lembik

or more specifically, what size NN do you need for that

Fatalize

If you don't overtrain it it will probably find patterns like "not pair",etc.

Lembik

@Fatalize I was just aiming for better than random not 100% :)

flawr

if you only have a finite number of inputs, just make it big enough

Fatalize

I'll give it a try...

Lembik

4:41 PM

I would like to

I have no gpu and no code :)

flawr

@Lembik thats already gonna happend if it just tells even from odd numbers

Lembik

I can fix one of those

@flawr I would already be impressed if it managed that to be honest

Fatalize

@Lembik there's no way it can't do that

flawr

A nn that just tells even numbers apart from odd ones?

Lembik

@Fatalize ok.. so how far do you think it could get? multiples of 3?

do you see what I mean? It could be interesting

flawr

4:42 PM

probably heavily depends on the encoding you use

Lembik

but I need some easy to play with NN code and a gpu

@flawr yes I am sure that is right.. which why I thought it could be fun

flawr

but I'm more interested in an NN that could actually take inputs of arbitrary sizes

Lembik

@flawr hmm... up to some limit I suppose?

flawr

without limit, as long as you have enough time

Lembik

there are recurrent neural networks but I don't really understand them

I feel we should use sat solvers more at ppcg... just saying :)

Dr Green Eggs and Iron Man

4:49 PM

How to generate a fractal in vim: qq<C-w>v<C-w>n@qq@q

flawr

@Lembik what are sat solvers?

Lembik

@flawr do you know the problem SAT? That is the task of finding a satisfying assignment in propositional logic.

flawr

No, I do not really know logic

let me ask wikipedia

Lembik

@flawr ok well it is the classic NP-complete problem

do you know what np-complete means?

Transcript for