Blaine reaches for the both of Anderson’s attention had been rumors splashed across a sliver of exposed flesh. His erection is bumping the mattress.‘So, tell me how much I missed this. The air in gasps, mouthing up the side, with the lemons life threw their way upstairs to Blaine’s a whole different feeling from a glowing and swelling even more at the buttons of Kurt’s voice, post - orgasm purr, is a wonderful feeling that one night when he tries to remember ) Blaine stares back, my stepmom Carole, cooing and prodding until Blaine is sweating and sated surrounded by these four huge Alphas. It cuts off quickly, and it’s thoughts a cow moos in the space that feels good, sweetheart, we, like, a little start that conversation. Blaine sets his messenger bag down and licked along Kurt’s hard cock starting to rock back against the pillow. Kurt nearly broke character and sank almost painfully hard squeeze. “ You got another one and then dropping his hands bunched up against Kurt.
I wrote a long-ish post over on DW about using a simple machine learning/natural language processing technique to write computer-generated fanfic and the hilarity that results from it. You can see some of the beautiful results above. Want to know how it’s done? Check out the link above.
I love you so much right now.
The reason I wrote up that post with bigrams is because that’s what I could get nltk to support out of the box with their example in the book, and it was the easiest to get up and running as a prototype. I would have used their ngram model which uses some more sophisticated tuning mechanisms on the probablilities if it weren’t kind of broken at the moment and still getting fixed.
Anyway, my point is that I wrote that post a whole 5 days ago, and I now have my code up and available (and even mostly documented!). It now supports arbitrary n values, so if you want to mess around with n values higher than 2 or 3, you can. I’ve settled at 3, because with 4, it has moments where it looks like it’s just copy/pasting words out of someone’s fic at times (like you said, constrained), but it is a lot more coherent. The computation shouldn’t be all that more expensive. It’s all on the same order of magnitude. I haven’t really benchmarked it, though.
cucumber-s
