The Nineteenth Byte - 2018-04-22 (page 3 of 3)

« first day (2640 days earlier) ← previous day next day → last day (2195 days later) »

4:11 PM

@mınxomaτ That's incredible, and slightly creepy

It's amazing how lifelike they look

TIL, the DGX-1 is only $150k. That's surprisingly cheap for what is essentially an SFF supercomputer.

all hail the video fake news era

I wait for the day when I can purchase my celebrity online persona

That seems like a waste of time and money.

Wait I think I just described the internet.

@mınxomaτ No, that's just the web.

4:20 PM

0

Q: The Top Ten Elements You Won't BELIEVE Are In This Array

AKA: Generate Clickbait From an Array. Given an array of integers, generate some cringe-worthy clickbait based on its arrangement and length: If it's 20 elements or less, you can make a Top X List. Parameters: length of the array. Prime numbers are celebrities, so anytime two of them are next ...

code-golf array-manipulation

@DJMcMayhem Wait 5 more years, and they'll climb out of the other side of the uncanny valley. (If people are still developing them.)

@wizzwizz4 Oh it's already way past uncanny valley. That's what's creepy about it. Knowing that computers can algorithmically create pictures that are so lifelike that I can't tell them apart from real humans

@DJMcMayhem You won't like it when we start using Super GPUs.

I mean Quantum Computers.

I'm pretending that Quantum Computers are like really parallel GPUs because I don't want to abandon conventional computers because I have an affinity for them.

I think modern day Pascal cards are pretty much super GPUs already

@DJMcMayhem How many cores?

If it's not easiest to write the number as ~$2^x$ that's not a Super GPU.

4:30 PM

@wizzwizz4 2560 CUDA cores @ 1607 MHz

I've never heard the term "Super GPU" before. Is that a real technical distinction?

@mınxomaτ generative adversarial networks are some evil satan shit

like, it's obvious why they work but do they have to work so well, it's freaky

@quartata Abstract for your next paper?

@DJMcMayhem I don't think so.

@mınxomaτ someone definitely should

anyways it's pretty cool how "continuous" the output is when you sample from them

like how they just interpolate between the inputs and you get a transition

@quartata Can't you use those to upgrade low frame-rate videos?

4:35 PM

that's not really what I meant

these images are made by taking the decoder half and randomly sampling the latent space

I wonder if GANs could be used for linguistic training rather than photo generation

what I thought was cool was that as you move from one point to another in the latent space (via interpolation, that's what they're doing) you get that fluid transition affect

@DJMcMayhem I've seen them used for text before

part of the trouble is handling sequences

You've changed your profile picture.

GANs can't really do discrete output, for one

@quartata Can you just chop up the continuous output? Some people try that with NNs.

4:38 PM

That and it would be really hard to make the computer evaluator be anywhere near as effective as a human interrogator

@DJMcMayhem that's not a problem so much

@wizzwizz4 not sure what you mean

@quartata If you want discrete output, calling this range of continuous "A" and this range "B".

If you've put them into a sensible order.

that doesn't work on a character level

if you mean word embeddings, then yeah sure

@quartata Why not? Conversations are interactive/adaptive

the trouble is still that GANs aren't recurrent

meaning they don't have a state as you go from word to word

you can feed in a fixed length sequence of words

I'm sure that something with 1D conv at the input layer could work with that

@DJMcMayhem these wouldn't really be used for conversations per se

4:43 PM

Ah. So they couldn't be used to make a lifelike conversation bot

stuff like GANs are called "generative" because they learn a probability distribution

you can then sample randomly from it

so you can generate random images or random sentences, but by itself it's not going to respond to something like you're envisioning

I was thinking you could use one to attempt to pass a Turing Test, but that's clearly the wrong application

well it is kinda funny

the idea behind the GAN is that they learn by having one network challenge the other

(that's the adversarial part)

@quartata Network 4725 or Network 0424! Which is better?

@quartata I think the #1 reason the Turing Test is so hard is because you can't use a human interrogator to run hundreds of generations of adapting

4:45 PM

in essence, one half of the network learns to create things, the other half learns to tell the difference between what it made and what is real

@DJMcMayhem Of course you can! Yahoo! Answers.

Eww

@Downgoat Somebody made a comment about goats.

@quartata How can both halves learn at the same time? I get how the generator can learn, but I'm not sure how the discriminator can learn while it's being fed false information

@Downgoat chat.stackexchange.com/transcript/message/44090318#44090318

4:49 PM

@DJMcMayhem at each step you generate some samples, you train the discriminator to classify the real images vs the samples, then you train the generator to try maximize the number of samples the discriminator marks as real

that's why they're adversarial

once the discriminator learns what the samples look like, the generator tries to change things to stump the current discriminator

and then you repeat with the new samples

so it's more similar to a turing test than you might think, just that both parts are automated

And I assume each generation is done with neural network evolution?

yeah, simple backprop for both

they are just normal neural networks

in the case of images they usually have convolutional layers though

@quartata Wikipedia pointed out that it's similar to a Turing Test, I didn't notice that on my own :P

oh, heh

@quartata Would an GAN made with a Convolutional Neural Network be able to generate news articles?

4:54 PM

Ooh, that's a fun idea

like I was saying earlier, I think maybe 1D convolution + a word embedding could do some text generation

let me look, I thought I remember an article

pretty much can just take a look at the related work here, there's several options

This is exactly what I love about TNB. There are people who know about so many different topics. I know very little about AI, and I've never heard of GAN before today, but I have a decent understanding after a short conversation

haven't read the whole article yet but it's co-authored by one of the original GAN folks so I'm sure it's Good

0

Q: DNA Transcription Python 3.x Code Failing "ACGTXXXCTTAA" test

This question may resemble this question but the problem is not exactly same and I was unable to find the answer there. I need help with my code correction in Python 3.x only. Therefore, before downvoting and deleting this post, if you could just help me with the problem that'll be helpful. A li...

@quartata just to verify, the discriminator is only trained on authentic data, right?

5:00 PM

it's trained on a classification task

it's supposed to output 1 on the authentic data and 0 on the generated data

so you do need some authentic data, it's supervised learning

Oh, that makes sense

(labeled authentic data I meant to say)

It would be really entertaining to train a GAN to generate clickbait articles

@quartata You have a new job. :-)

I'm about to move, and I'm looking into internet providers: One of them has data caps, so I was checking to see how much we use: my modem says 12GB in the last 24 hours, but that seems really high. We do stream, and generally have music going throughout the day, so I'm looking for a reality check

I don't know how much data streaming actually takes (and I don't trust ISP's suggestions), which is why I'm asking here if that seems right

5:12 PM

Music will probably be negligible, movies will take up quite a bit. Not sure if 12 GB is realistic though

For example, I listen to music all the time, and my Spotify Mobile data bandwidth over the last month is 55 MB. Obviously most of the time I use WiFi rather than data, but still

12 GB sounds way too high

unless you've downloaded something recently

@Downgoat what's on the DO droplet that you used to run the TF2 server on?

nope. I did a couple days ago, but not in the past 24 hours

@quartata Either debian or centos

no I mean, is there anything running on it currently

CMC: Given N-window slices of an array and N, produce the original array. E.g.:

[[1, 9, 3], [9, 3, 3], [3, 3, 7], [3, 7, 4]] , 3
=>
[1, 9, 3, 3, 7, 4]

5:17 PM

I'm probably going to be looking for people who can get me a SourceTV relay

(not the actual game, that'll be on my own EC2 probably)

just an STV relay for spectators

@ConorO'Brien Can we assume that all pieces of the array are a single character? (for vim)

@DJMcMayhem sure

@quartata yeah it’s running all vihan.org stuff except axtell

@ConorO'Brien lambda l:i[0]+[i[-1]for i in l[0:]]

@Downgoat what's your free resources looking like

can I steal your shit for like 2 weeks

5:20 PM

@ConorO'Brien Haskell 28 Try it online!

I just don't want to have Nginx, STV and the bots running on one thing. I'd like to spread it out a bit

make the forward facing stuff separate from where the bots run

~~and also i'm not spending money on a ppcg challenge~~

@ConorO'Brien V, 7 bytes: òjklDgJ Technically, you don't actually need to know N

yup

This is sometime in the future by the way, not like right now

It's for a KoTH

Haskell 20 Try it online!

5:23 PM

@WhatWizard 19 bytes With 8.4.2

Nice, what does (<>) do?

oh yeah I remember that tip

@WhatWizard It is mappend. For functions, it mappends their results

You can now use <> instead of mappend, it's now (with GHC 8.4.1) part of the Prelude. — BMO Mar 10 at 15:50

ninja'd

Ok cool.

5:24 PM

@ConorO'Brien Canvas, 5 bytes

Are Haskell monoids what I think they are?

Presumably, although the new change is that they are to be a subclass of Semigroups

So what does <> do in the more general context of monoids?

here is a nice use of for quining

@ConorO'Brien Can we assume that there will always be atleast 2 windows? Can we assume that all integers are positive? (I'm doing it in brain-flak)

5:27 PM

(<>) :: Modoid a => a -> a -> a

instance Monoid b => Monoid (a -> b)

Ok so it is the monoid action. And I guess functions are defined with a particular action

The main use of <> in the Prelude is ++

Oh yeah, I bet that would be useful in some restricted source contexts.

instance Semigroup b => Semigroup (a -> b) where
f <> g = \x -> f x <> g x
stimes n f e = stimes n (f e)

How does Haskell enforce that exists identity?

5:36 PM

It doesn't, in the same way it can't enforce the laws of other structures, like monads

For monoids, you define mempty, but you could define it to be anything

@ConorO'Brien Brain-flak, 88 bytes: ([({})()][]){{}(({}<({}<>)<>>[()])<{({}[()]<{}>)}{}>())([({})()][])}{}{}<>{({}‌<>)<>}{}<>

@WhatWizard While I was writing that, I accidentally came up with some painfully dumb and non-golfy things. For example, I wrote ([({})][][()]) and...

{({}<>)<>}{}
<>
{({}<>)<>}{}
<>

Erik the Outgolfer

@ConorO'Brien you don't need N as input...

Jelly, 4 bytes: Ḣ€;Ṫ

5:52 PM

I strongly feel The Top Ten Elements You Won't BELIEVE Are In This Array needs closing until further clarification (see my comments) the author seems to have posted and not responded to these questions (yet has been active since the first two of my comments were posted). If answers arrive assuming certain interpretations it will become harder to clean up.

6:04 PM

...most problematic concerns addressed now :)

6:15 PM

@EriktheOutgolfer yup

3 hours later…

9:06 PM

user image

this is my pet peeve

@JonathanAllan needs one more reopen

@quartata I already cast

New Sandboxed Posts

0

A: Sandbox for Proposed Challenges

Single-Digit Representations of Natural Numbers Introduction Dr. Inder Taneja is a Mathematics Professor at the Federal University of Santa Catarina who wrote and published "Single Digit Representations of Natural Numbers". In this paper, he describes each natural number from 101 to 1000 as in ...

@quartata Done

9:24 PM

@quartata around 40% free but no

@ConorO'Brien Stax, 4 bytes: {H+k

that's ok, I thought you has less stuff on there

(i still consider it a miracle the ppcg tf server ran)

although STV relays aren't nearly as intensive

fwiw I didn't mean disk space, I meant RAM

that was the problem when we ran anything on their last time, your wordpress instance took most of the ram

@Downgoat wait, so where is Axtell running? thought you had moved it from Mego's dedicated box

what about zeros inside though?

@quartata we are running it on a different DO box

@JonathanAllan Whoops

Fixing

9:33 PM

@quartata yeah, I switched from wordpress -> Ghost and now droplet has 2GB of ram instead of 512 but I'm also running a couple personal things from that server too though

have you tried Jekyll

I was thinking about that but I'll have to spend some time configuring stylesheets etc. that I have for Ghost rn

oh yeah I dont know much about its themes

@ConorO'Brien APL (Dyalog Unicode), 8 bytes ⊃¨,1↓∘⊃⌽

Does golfing code without paying attention to all the beer-drinking drunk people on the train yelling and singing count as do X without Y?

7

Anonymous

@quartata It's on my droplet

9:42 PM

@Adám It's a shame that ,/⍣¯1 doesn't work

@Adám yes. definitely. ugh...

@H.PWiz Yeah, I tried that. Want to email support@dyalog.com about it?

Not really

@H.PWiz OK, then I'll file an RFE.

@H.PWiz Filed as issue 15761.

Cool, hopefully we will get inverses for f/ as a dyadic function.

10:48 PM

Actuall production code I work with:

public static void Main(string[] args) {
	// Do not remove this try/catch! The program will crash instantly.
	// No one knows why this works so don't touch it.
	try {
		//...
	} catch (ArrayIndexOutOfBoundsException WTF) {
		Console.Error.WriteLine(WTF);
	}
}

what.

I removed the try/catch to see if it actually happened, and yep, immediatly crashed with ArrayIndexOutOfBoundsException.

The best part is the stack trace indicated it was thrown from a spot where no array access happens.

Is the actual work done in other threads?

It's a single threaded application, which is why no one has any idea what's going on.

Our best guess is that one of the libraries we're using is using threads and we don't know about it.

@H.PWiz Oh, right, dyadic f/ is invertible for value-preserving functions too. I'll add that to the issue.

10:59 PM

@Pavel That wouldn't cause that. I think what feersum means is that your main thread exits immediately if it catches that and just logs it and returns from main

I'm not convinced your application, like, works

@quartata It totally does.

IIRC the JVM only exits when all threads that do not register themselves as "daemon" threads have returned.

I don't remember what happens when main throws an exception, but I was hypothesizing that that would cause the program to terminate.

Like it makes a window and you can click on things in the window and they do the things they're supposed it.

@feersum this looks like C#, but I'm pretty sure this is also true

While a normal return from main does not.

11:01 PM

It's C#, yeah.

hahahaha oops

@Pavel Ah, so it is doing things in separate threads then

That explains it

And it seems to be some library code that's failing, because we don't use threads.

Java has an exception with the exact same name I believe.

Threads wouldn't cause an exception in the main thread like that

It wouldn't get caught in your setup and would say which thread instead

11:03 PM

Ok the actuall exception is IndexOutOfRangeException since I was retyping what was written and I forgot what the actuall exception is called

My idea was that the exception occurs in the main thread, but the catch prevents the other threads from being killed.

@feersum yeah

There's also an [STAThread] attribute to Main. I don't know what it's for.

CMC (or main?): Given a list of pairwise sums, return an original list which would give those pairwise sums. E.g. 4 5 5 6 14 → 2 2 3 2 4 10 or 3 1 4 1 5 9. (There are obviously infinitely many valid solutions.)

Except that it stands for SingleThreadedAppartment

@feersum The thread doesn't stop or anything, it goes on to repeatedly print out stacktraces of the same exception until you exit.

11:05 PM

oh god no

is this WinForms

It is not.

Well way to not tell us the most important information!

...are you sure

Definitely.

@Pavel even with debugging on?

11:06 PM

Yes

Oh no I GTG

because that's where STA comes from, it's a load of COM compatibility bullshit

o/

I kind of wish we had stuff like this where I work.

so then it is threading?

Everyone is too competent and it's boring.

4

11:09 PM

0

Q: Decode the string

This is my first challenge on ppcg! Input A string consisting of two different ascii characters. For example ABAABBAAAAAABBAAABAABBAABA Challenge The task is to decode this string following these rules: Skip the first two characters Split the rest of the string into groups of 8 characters...

code-golf string

O_o

looks pretty straightforward to me

SE is probably blocked in China

@Adám f(a){return a;}, because 4 5 6 7 + 0 0 0 0 == 4 5 6 7?

@betseg adjacent pairs

@betseg No, not that kind of pairwise.

11:14 PM

oh im stupid

@Adám: scanr(-)1

@BMO Whoa, a built-in.

Kind of yes, I guess APL has something similar as well?

@BMO What does =<< do?

It's the infamous bind for Monads

11:21 PM

Huh

In that case it's the IO monad, so it would type as (>>=) :: IO a -> (a -> IO b) -> IO b

oh wow, haskell's site is so good

@BMO What does scanr do?

@betseg Hoogle is the best

Where IO a would be a string and a -> IO b is the print . f . read

11:23 PM

@Pavel haskell.org/hoogle/?hoogle=%3D%3C%3C :P

@Adám: source

It basically takes a binary function and does a pass over the list using the last generated element and the one in the list and accumulates them in a new list, starting (in that case) with 1

@Adám It's quite like \

It's like C#'s Aggregate or Python's reduce I think

@H.PWiz \?

in APL

For example scanr(+)0[1,2,3] = [0,1,3,6] (no it doesn't)

11:28 PM

@H.PWiz Sure, I just meant the markdown: backtick, backtick, backslash, backtick, backtick.

@Pavel That would be foldr (or foldl), reduce returns a single element whereas the scan accumulates the result, however you could use reduce to achieve the same

@Pavel nope

@Pavel more like deltas with an arbitrary function instead of only supporting(-)

@H.PWiz Right, I gathered that much, so how would you write this in APL?

@Adám GIve me a feew minutes ... It's quite late

11:40 PM

Why did JavaScript take the worst parts of Java and none of the good ones T_T

who doesn't love 25 lines of imports at the beginning of files

@Downgoat What good parts of Java would you like to see in JS

@Pavel real classes

@Downgoat So use TypeScript

that is half-baked glue that attempts to maintain perfect JS interoperability but fails miserably.

It really doesn't

11:45 PM

yes it dose

TypeScript works fine, I've never had problems with it

I didn't use it all that much, but it seemed ok.

one of the primary points of having a type system is to avoid bugs caused by passing the wrong thing. If you try to implement polymorphism with union types though you don't really get that anymore though

It's still better than just JS

true, but then I don't get optional chaining and other babel features

Magic Octopus Urn

CoffeeScript :D?

11:58 PM

RIP CoffeScript.

Magic Octopus Urn

Oh, is it ded?

coffeescript is ancient and lacks new ES features

Magic Octopus Urn

Haven't used in 4 years.

Oh, he ded.

It was a great thing until ES finally caught up to it.

« first day (2640 days earlier) ← previous day next day → last day (2195 days later) »

Transcript for

Apr '1822

The Nineteenth Byte

The Nineteenth Byte: General discussion for codegolf.stackexc...

join this room
about this room

00:00

06:00

12:00

18:00

all times are UTC

site design / logo © 2024 Stack Exchange Inc; legal

mobile