THE FUTURE IS HERE wrote:
@habs
Can you ELI5 your process of creating?
Sure. Basically, when trackbot reads posts, even ones that don't call it, it archives them all in a mySQL database which I used for this. For a general explanation of Markov chains you can read this reddit post:
https://www.reddit.com/r/Python/comments/2ife6d/pykov_a_tiny_python_module_on_finite_regular/cl3bybj/Personally I'm currently using the markovify python3 library. Because my corpus is way to large (~100MB of pure text) to use all at once, every 20 minutes it generates a new corpus with 1000 posts, 5000 titles and 5000 usernames, all randomly selected from my DB which goes back to August 2016. My script to generate the models is here and that automatically gets run every 20 minutes:
http://ix.io/1RBw/pyThen every time you load the page, it uses the current models to create the posts and puts it in a 'letsrun-like' page template to get the feeling of being on the boards. Code for that is here:
http://ix.io/1RBx/py