Read this if you're procedurally generating wilderness maps!https://twitter.com/AboundDragons/status/1092449866008018945 …
-
-
Replying to @redblobgames
I like this a lot. I'm used to thinking about this from a slightly different point of view: given a physical system (often abstracted to something simple like a graph), you'd like to understand the "typical" state. But often it's hard to sample uniformly from the true ... 1/n
1 reply 0 retweets 3 likes -
Replying to @blockspins @redblobgames
"possibility space" so you come up with approximations which give you instances from a kind of "generative space". There's lots of beautiful phenomena in stat mech and probability in this area, like the "arctic circle theorem" (image by
@JimPropp from http://faculty.uml.edu/jpropp/tiling/www/intro.html …) 2/npic.twitter.com/wD1hxfYVwO
1 reply 2 retweets 9 likes -
Ahh... there's way too much I could say so I'll stop. Anyways, it's fun to see other communities exploring some of my favorite math / science ideas in a different language and with different motivations! 3/3
1 reply 0 retweets 1 like -
Interesting! I wonder if some of this theory would help to sample random metrical verse -- think of words as tiling a verse template, but successive words have different affinities, and so do ends of lines for rhyming. I played with that years ago in a brute-force way.
1 reply 0 retweets 1 like -
I found your Sonnetron http://wry.me/sonnetron/ , it's so cool! One direction to explore: given an "energy" function that penalizes verses which slightly miss the template, and a choice of "steps" to take to mutate verses, there's a natural MCMC sampler: https://en.wikipedia.org/wiki/Markov_chain_Monte_Carlo …
1 reply 0 retweets 0 likes -
Thanks! I considered trying something like that (though I don't understand much about MCMC) but there's something charming about backtracking, that it looks a bit like a sped-up movie of a human drafting an essay.
1 reply 0 retweets 1 like -
Replying to @abecedarius @blockspins and
But I think you're right that that's a more promising approach to generating more ambitious kinds of output.
1 reply 0 retweets 0 likes
The one-line takeaway might be "random walks in 'possibility space' are sampling algorithms" (though not nec. vice versa). Your backtracking algorithm and MCMC are just two different choices of walks. Now if we want we can think about sampling from sampling algorithms...
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.