markov

Markov

A simple Markov chain generator. Originally created for the Language Modeling lecture in the Introduction to Synthetic Media class at ITP/NYU.

Download

Binary releases can be downloaded from the releases page. Be sure to unzip your file, and put the markov binary somewhere in your $PATH (like inside /usr/local/bin/).

Usage

# Download the UCI News Aggregator Dataset (400K news headlines) as a sample corpus
wget https://github.com/brannondorsey/markov/releases/download/v0.1.0/uci-news-aggregator-dataset.txt

# Build and cache an n-gram frequency histogram, then use it to generate text
markov --corpus uci-news-aggregator-dataset.txt --n-gram-length 3 --prompt "For the first time in a decade"

Below is a summary of the full usage of the markov command.

Usage of markov:
  -i, --corpus string        The input corpus to build the n-gram histogram with.
  -h, --help                 Show this screen.
  -l, --lowercase            Convert text to lowercase. Lowers the complexity of the sampling task, and may produce better results depending on the corpus.
  -c, --max-characters int   The maximum number of characters to generate. Fewer characters may be generated if the sequence encounters an n-gram that has no next n-grams in the dataset. (default 1000)
  -n, --n-gram-length int    The number of characters to use for each n-gram. (default 3)
  -p, --prompt string        The prompt to (optional). (default "hello")