A game built on Markov chains
This app requires Python 3.6+ and Node.js 6+.
To make it work, you first need to build the Markov chain, which is not included in source control because it is very large. The file markov/raw_tweets.txt is the raw corpus for the chain. It is a dump of several hundred thousand random English-language tweets, with retweets removed.
pip install -r requirements.txt
python markov/sanitizer.py # Sanitize the raw tweets to normalize case, remove URLs and @mentions, etc.
python markov/markov_builder.py # Build the Markov chains and dump them to JSON
This will generate two files, markov.json and markov_mini.json. The former is intended for production use, but it takes a long time to load because it is so large, so the latter is a much smaller version of the chain. It is intended for development but provides much less interesting gameplay.
Build client-side assets and run the app:
npm install
npm run build
python app_init.py
For production usage there are also systemd and nginx config files.
If you want to add more tweets to the markov chain you can just run the tweet loader, which will add raw tweets to the raw_tweets.txt file.
Set the TWITTER_CONSUMER_KEY, TWITTER_CONSUMER_SECRET, TWITTER_ACCESS_TOKEN, and TWITTER_ACCESS_TOKEN_SECRET environment variables and then run:
python markov/tweet_loader.py