|
I did this but with YouTube comments the output was real hosed up
|
# ¿ Feb 27, 2016 07:13 |
|
|
# ¿ May 6, 2024 09:00 |
|
https://twitter.com/deepdrumpf
|
# ¿ Mar 5, 2016 15:40 |
|
Doc Block posted:Best one IMHO it's loving beautiful. better than actual weird Twitter humans
|
# ¿ Mar 25, 2016 02:07 |
|
just installed fedora 23 on some desktop that was gifted to me proceeded to instlal torch rnn and the thousands of dependencies, cuda, nvidia driver, etc holy loving poo poo how does anyone with a life use linux. i had to gently caress with source headers to even get cutorch to compile
|
# ¿ Apr 17, 2016 06:41 |
|
i shouldve put my better geforce in this pos, although im sure at this point if i swapped gpus itd totally brick. training an lstm on ulysses is slow
|
# ¿ Apr 17, 2016 07:00 |
|
to anyone using torch-rnn, try using code:
|
# ¿ Apr 17, 2016 17:22 |
|
atomicthumbs posted:Sampling is on by default, if you do sample 0 it uses argmax and is boring. Temperature is 1.0 by default and I usually find something between 0.3 and 0.7 works decently without just being gibberish lol i came back to post this after reading sample.lua. how does this thing handle whitespace? just like any other char? the example where the char-rnn guy did linux kernel code was amazing and i want to try it on some other code
|
# ¿ Apr 17, 2016 21:20 |
|
training rnn on a huge enterprisey java codebase, going to take like 8 hours but the output from early checkpoints is p goodJava code:
|
# ¿ Apr 23, 2016 15:49 |
|
eh well it crashed with a memory error after a few hours but not before giving me this gemJava code:
|
# ¿ Apr 23, 2016 18:08 |
|
terminator scriptcode:
|
# ¿ Apr 23, 2016 19:47 |
|
Pumpy Dumper posted:ive been feeding mine seed words. i still havent implemented automatic picking because i have to implement error checking first. some words when inserted, even though they have entried in the corpus, kick out key errors. I suspect this is due to markovify's built in sentence checker? not sure if you're already doing this for seeding but you might be able to use NTLK to get some of the subject words from a post the bot is replying to. then again maybe bad idea because the bots would converge to Pittsburgh mono posting feedback loop
|
# ¿ Apr 24, 2016 15:40 |
|
kalstrams posted:they would converge on *jerking motion* ... eventually
|
# ¿ Apr 24, 2016 15:44 |
|
did you scrape the seinfeld scripts yourself? i tried this with like a handful of episodes but was too lazy to get all of them. 5mb should be enough
|
# ¿ May 3, 2016 15:10 |
|
you need a GPU or a lot of patience
|
# ¿ May 3, 2016 22:14 |
|
|
# ¿ May 6, 2024 09:00 |
|
goddamn please post the seinfeld corpus
|
# ¿ May 6, 2016 20:32 |