Markov bot instead of ELIZA clone (easier, faster to create)

- Lightning:

- https://finalfantasy.fandom.com/wiki/Lightning_Returns:_Final_Fantasy_XIII_script

- http://ffxiiitranscript.blogspot.com/

- Balalaika:

- https://www.opensubtitles.org/en/search2/sublanguageid-all/moviename-black+lagoon+

- https://www.quotes.net/movies/black_lagoon_100757

- https://tvtropes.org/pmwiki/pmwiki.php/Quotes/BlackLagoon

- https://www.imdb.com/title/tt0962826/quotes/

- https://duckduckgo.com/?q=black+lagoon+balalaika+quotes+imdb&t=ffab&ia=web

- Yandere:

- yandere ASMR scripts(?)

- My discord logs

- My writing

- My quotes.org file

- Douglas Adams books

- Zero Punctuation transcriptions

- Izaya Orihara quotes

- Hazama quotes

- Terumi quotes

- n-gate

- Hackernews comments

- specify how input files work (no punctuation, lines of sentences)

- discuss why interact() only does single-word input

-h: show help

-t: train only (outputting to "weights.json" if the output file is not specified)

-r: run only (using "weights.json" if the weight file is not specified) - read from stdin

-o: specify output file

-w: specify weight file

-u: "unweighted" training; weights only contain distinct words, with no information about how often the occur

-l: specify minimum response length

Right now, quickbot only handles single-word activation functions, and only ever will, as far as I can see. Why? Because there're a million ways to run a markov bot, and I don't want to handle all of them. Use another script that munges the incoming message and pipes in or something. It'll be easier. In the meantime, here are some ways of running it (IRC-style) that I though of:

Use the last word in the input

Use the first/last word in the input that's in the vocab

Use a random word from the input that's in the vocab

Use a word that's in vocab and has the most associated words

Do multiple sentences using each word in vocab