Have duckbot mimic a given user

[twentylemon]

We have a database now. Let's use it.

For each user, we can break down messages as they come in to build their transition matrix over time, storing the matrix in the database. So, if a human says I like food very much, we could store the n-grams I like food, like food very and food very much, or whatever n we want. Naturally larger n builds better sentences, but our data is likely to be fewer words in general, plus we do have to store it all.

Then, with those transition tables, we could have a command to mimic a given user.

Human: !markov @AnotherHuman DuckBot: Some markov chain nonsense using AnotherHuman data!

We could also have additional arguments to !markov, like a starting point for the sentence to generate, or how many words to generate.

[Chippers255]

I would also like to see the idea of seeding the markov chain with the !markov command. Such as !markov @human guys, today I learned that...

[twentylemon]

nltk has ngrams which we can use, we already use nltk as a dependency

http://www.nltk.org/api/nltk.html#nltk.util.ngrams

[twentylemon]

Also relevant reading: https://softwareengineering.stackexchange.com/questions/64288/storing-n-gram-data

[twentylemon]

https://github.com/Chippers255/channel_police/blob/main/extract.py is some code to fetch channel history, for the initial building of the markov transition matrices

[twentylemon]

See #701 for a preferable approach to markov chains

[Chippers255]

This has a lot of potential.