Tumgik
#also you can generate your own docbot tweets at any time via the pinned!
botm77 · 2 years
Note
Have you made any changes to docbot since you started this? (I don't know coding very well (at all) so this may be a dumb dumb ask)
this is not a dumb ask! botm77's content is produced via the absolutely ancient markov chain generator contained in the spreadsheet in his pinned post, which I have used for various bot_ebooks projects since 2014. unfortunately this means all his "tweets" are capped at 140 characters (the old limit), and I've made futile attempts to adjust the script without any luck. whenever I feel like it or I see Doc tweeted new words I want in the corpus, I run a new TAGS pull (info in spreadsheet) and filter out RTs and uninteresting replies.
(incidentally, this is why he used to post a lot more about nuclear waste - before I cleaned up the source material, doc arguing about nuclear power plants accounted for 1-5% easily of the tweets I could pull. this tendency has been eliminated through trial and error.)
a very small number of posts were created using some markov chain generators I found on the web as an experiment, but not many and I wasn't happy with the results vs. the spreadsheet.
my dream, since tumblr is unfriendly to automation and I have to queue everything manually, is to get a more advanced AI program such as textgenrnnn going. unfortunately, the google collab doc I have for textgen is no longer functional and i am too busy to learn enough python to troubleshoot. markov chains are good for simple ebooks, but since this one has an outdated character limit and also the corpus is not friendly to longer string, the quality of outputs are limited in terms of novelty if nothing else; you see a lot of structures like "just got this is insane" where you can clearly see the two logical strings that got stitched together by the algorithm; I also curate out a lot of the repeated phrases that become less interesting when they're used over and over by the generator.
in the future, significant changes to the code I use for posts will probably be at least updated in the pinned post here! but I am also capricious and extremely lazy, so do not hold your breath.
7 notes · View notes