itt markov bots: The aging millennials speak in their crude dildo pidgin.

The Something Awful Forums > Discussion > Serious Hardware/Software Crap > YOSPOS > itt markov bots: The aging millennials speak in their crude dildo pidgin.

vodkat: Jun 30, 2012; cannot legally be sold as vodka

this is v. cool and I am defiantly stealing this idea as my next idiot spare time project. I'm going to make a little zoo of markov bots and pretend that real people still use twitter :nsa:

# ¿ Feb 14, 2016 02:59

Adbot: ADBOT LOVES YOU

# ¿ Apr 27, 2024 19:00

vodkat: Jun 30, 2012; cannot legally be sold as vodka

kalstrams posted:

https://twitter.com/markov_polov/status/699239206145556481

fuckin owned

# ¿ Feb 15, 2016 19:31

vodkat: Jun 30, 2012; cannot legally be sold as vodka

gonna try and write a markov bot in python tonight.

will report back when i inevitably fail.

# ¿ Feb 18, 2016 19:10

vodkat: Jun 30, 2012; cannot legally be sold as vodka

My own dumb markov bot is almost ready to go. Unlike trig's the tweets are going to be drawn from people my bot follows, so does anyone have suggestions of twitter accounts he should follow?

# ¿ Feb 20, 2016 18:21

vodkat: Jun 30, 2012; cannot legally be sold as vodka

Trig Discipline posted:

oooh nice i was sorta thinking about that but i'm headed to japan shortly and

that's a super versatile way of doing stuff because you can totally change what kind of bot it is just by changing who it follows. i'd say make a few with different themes.

That was the plan once its up and running but it took my a while to get the whole thing running because python doesn't have a nice library to use for making markov chains (nltk used to have one but it got depreciated). Luckily I found someone else code that I could steal that seems to work well enough.

For a super hardcore markov bot you make a per character neutral network like this guy http://karpathy.github.io/2015/05/21/rnn-effectiveness/ but that is way beyond my lovely programming skills or my MacBooks integrated graphics.

# ¿ Feb 20, 2016 18:34

vodkat: Jun 30, 2012; cannot legally be sold as vodka

And our first tweet is up

https://twitter.com/markov_botman/status/701097785026805762

# ¿ Feb 20, 2016 18:35

vodkat: Jun 30, 2012; cannot legally be sold as vodka

a sinister turn already

https://twitter.com/markov_botman/status/701102225200906242

I should follow some other people that aren't echi

# ¿ Feb 20, 2016 18:55

vodkat: Jun 30, 2012; cannot legally be sold as vodka

https://twitter.com/markov_botman/status/701141137541746688

# ¿ Feb 20, 2016 21:28

vodkat: Jun 30, 2012; cannot legally be sold as vodka

I made a huge corpus (2 million +) word corpus for markov botman from stupid silicon valley VC but he's still not generating very good tweets. I'm gonna try and write a better tweet generator over the next few days and report back with the results.

# ¿ Feb 23, 2016 00:13

vodkat: Jun 30, 2012; cannot legally be sold as vodka

coffeetable posted:

post the file

https://www.dropbox.com/s/rzzrbzi8wl104ks/Archive.zip?dl=0

lovely script also included

# ¿ Feb 25, 2016 01:28

vodkat: Jun 30, 2012; cannot legally be sold as vodka

coffeetable posted:

it's an rnn, not a mc. mc wouldn't be able to indent things consistently (context isn't long enough)

see

http://karpathy.github.io/2015/05/21/rnn-effectiveness/
https://github.com/jcjohnson/torch-rnn

I'm playing around with these at the moment but I still haven't got anything I'm happy with. Charter rnn seems to work fine but takes ages to run since I don't have a gpu. Torch-rnn is much faster and efficient but the output so far has mostly been unusable.

# ¿ Mar 6, 2016 14:29

vodkat: Jun 30, 2012; cannot legally be sold as vodka

atomicthumbs posted:

related: someone please tell me there's a way to scrape more than 3200 tweets at a time off someone's timeline, because god dammit twitter why cripple your API

i think thats the limit, but you might be able to go further back by using the last tweet_id

also what are you using for your neutral network? I got char-rnn up and running but it was very slow without a gpu, i switched to torch-rnn which was much faster but just kept outputting poo poo, so I kinda gave up on this project for now.

# ¿ Mar 28, 2016 20:09

vodkat: Jun 30, 2012; cannot legally be sold as vodka

atomicthumbs posted:

i think i figured it out: torch-rnn's temperature defaults to 1 when I was using a random range of 0.3-0.7 earlier

here's top gun at 0.2

oh cool thanks for figuring out what the problem was, I'll try firing torch-rnn again later.

# ¿ Apr 5, 2016 16:41

vodkat: Jun 30, 2012; cannot legally be sold as vodka

Pumpy Dumper posted:

anybody got a good resource for making a python markov bot?

I stole my core markov functions from this guys code https://github.com/alexhwoods/alexhwoods.com/blob/master/Machine%20Learning/Markov%20Chains/Generate%20Text/Generate%20Text.ipynb, it fine but fairly simple. I didn't really develop it much more, instead I spent most of my time loving around with the twitter api. You could also write some additional stuff to check the output ect to see if its actually a good tweet.

Just fyi lots of answers about markovs in python reference nltk packages but these have all been depreciated and I didn't find anything else that would replicate the same functionality.

# ¿ Apr 13, 2016 04:42

vodkat: Jun 30, 2012; cannot legally be sold as vodka

atomicthumbs posted:

it's all neural these days, Mark Ov just didn't have the computing power to implement it

same, maybe in a few weeks I'll try running some proper nn in the :yaycloud:

# ¿ Apr 13, 2016 13:46

vodkat: Jun 30, 2012; cannot legally be sold as vodka

ChickenOfTomorrow posted:

how long before graph has to ban all of them

even that YOSMas 2014 posting bot that never got to stretch his wings

couldn't we just have a pet bots thread and some rules to limit excessive posting?

# ¿ Apr 19, 2016 18:48

vodkat: Jun 30, 2012; cannot legally be sold as vodka

yo thats a really neat idea. you might also want to run the corpus through nlkt stopwords to easily get rid of all the "and, but, to"'s etc and be left with words that might produce something interesting.

I'd love to try making GBS threads up the new thread since I already made a lovely twitter bot but I'm really busy this week and on holiday next week, by which time I expect you yospos geniuses to have made a tori bot that is actually indistinguishable from the real thing.

# ¿ Apr 20, 2016 18:33

vodkat: Jun 30, 2012; cannot legally be sold as vodka

Citizen Tayne posted:

I don't do any Donald Trump things.

Make Pittsburgh great again!

# ¿ Apr 25, 2016 22:30

vodkat: Jun 30, 2012; cannot legally be sold as vodka

Al! posted:

you were talking about how great Pittsburgh was since before everyone started talking about how great Pittsburgh was.

my wife to me the other day: "hey did you hear Pittsburgh was called the most livable city?"

it sounds like al! is the first one to completely replace his posting with a markov bot

# ¿ Apr 25, 2016 23:12

vodkat: Jun 30, 2012; cannot legally be sold as vodka

I'm Lovandie

# ¿ May 9, 2016 13:05

vodkat: Jun 30, 2012; cannot legally be sold as vodka

Powaqoatse posted:

btw i wonder if you could train a neural network on those guitar hero/rockstar/etc audio tracks

like use the [mixed track] => [original separate tracks] as ground truth so as to teach it how to separated any given track. probably too complex w current processing power & algorithms. would be pretty cool tho

you might have some success training a charter level nn with guitar tabs? It would be easy to scrape together a large corpus but it might be trickier to get it to do vertical lines? idk but it could be cool and there is probably a script out there can turn tabs into basic music.

# ¿ Jan 23, 2017 19:19

vodkat: Jun 30, 2012; cannot legally be sold as vodka

How are you training these bots to be so good? I assume you moved beyond Markov chains to some sort of neural net wizardry?

# ¿ Jan 24, 2017 15:24

vodkat: Jun 30, 2012; cannot legally be sold as vodka

thinking of trying to spin up a new twitter bot over the weekend, anyone got some good idea's for original source material?

# ¿ Jan 26, 2017 11:15

vodkat: Jun 30, 2012; cannot legally be sold as vodka

Powaqoatse posted:

i actually had to think for half a second

pooping when you gotta poop feels way better than listening to music, but then there's the forever qualifier

e: also yeah very nice clean breakdown for some reason

If you chose music would also have to give up posting

# ¿ Mar 19, 2017 02:14

vodkat: Jun 30, 2012; cannot legally be sold as vodka

Crusader posted:

on a similar note making machines hallucinate is kinda amusing:

https://arxiv.org/abs/1610.08401

https://www.youtube.com/watch?v=jhOu5yhe0rc

cool af and i look forward to defending myself from our robot overlords by being secretly wavvy

# ¿ May 8, 2017 00:43

vodkat: Jun 30, 2012; cannot legally be sold as vodka

Sagebrush posted:

if i'm on a date and the girl can't quote the three laws i'm out. i don't wanna potentially marry someone who's got such a blas� attitude to robot safety

i wanna create a nn tinder bot and see how many dates i get

# ¿ Jun 20, 2017 01:48

vodkat: Jun 30, 2012; cannot legally be sold as vodka

comedy option create a grindr bot and get a 100% success rate

# ¿ Jun 20, 2017 01:49

vodkat: Jun 30, 2012; cannot legally be sold as vodka

Powaqoatse posted:

I trained an A.I. to generate British placenames

https://medium.com/@hondanhon/i-trained-a-neural-net-to-generate-british-placenames-9460e907e4e9

dont go too near Stroby�s Hill or youll go blind, the old wives tale goes. feh, hogwash! i say

Most of these are real places and I've even been to some of them. You won't fool me that easily :colbert:

# ¿ Jul 23, 2017 23:44

vodkat: Jun 30, 2012; cannot legally be sold as vodka

Krankenstyle posted:

loling that you got a new username but are still stuck with that old red text

# ¿ Jan 20, 2018 22:12

Adbot: ADBOT LOVES YOU

# ¿ Apr 27, 2024 19:00

vodkat: Jun 30, 2012; cannot legally be sold as vodka

Improbable Lobster posted:

i'm gonna use that deepfake thing to recast movies

would make for a really good fan edit of being John Malkovich

# ¿ Feb 7, 2018 16:37

The Something Awful Forums > Discussion > Serious Hardware/Software Crap > YOSPOS > itt markov bots: The aging millennials speak in their crude dildo pidgin.