markov_str

markov_str is a fast and memory efficient Markov Chain implementation, optimized for text generation.

Features

• User can specify what regex they want to use for tokenization.
• MarkovChain::generate and MarkovChain::generate_start functions both take RngCore instead of using rand::thread_rng().
• Strings are interned for faster training and less memory usage.
• Serialization and deserialization with serde, when serialize flag is used.

Example

let training_path = "data";

// Gets the paths of evey file and directory in the training_path.
let tpaths = fs::read_dir(training_path)
	.unwrap_or_else(|_| panic!("Can't read files from: {}", training_path));

// Only the files remain
let files = tpaths
	.filter_map(|f| f.ok())
	.filter(|f| match f.file_type() {
		Err(_) => false,
		Ok(f) => f.is_file(),
	});

// Reads every file into a string
let contents = files.filter_map(|f| read_to_string(f.path()).ok());

// Creating the Markov Chain
let markov_chain = contents.fold(
	MarkovChain::with_capacity(2, 8_000_000, Regex::new(WORD_REGEX).unwrap()),
	|mut a, s| {
		a.add_text(&s);
		a
	},
);

// Number of tokens
println!("{}", markov_chain.len());

// Generation
for _ in 0..10 {
	println!("{}", markov_chain.generate_start("among the       ", 25).unwrap());
}

This example is taken from the examples/main.rs, you can run it by:

./get_data.sh
cargo run --release --example=main

./get_data.sh will download the first 200 books from Project Gutenberg, which totals up to more than 100MBs of text.

Changelog

Changelog

License

markov_str is licensed under the MIT license. Feel free to fork and use however you like.

Contributing

Feel free to open issues and pull requests. If you want to help with what I am currently working on, take a look at the Stuff left to do section.

Stuff left to do

• Multithreading support
• Arena for ChainItems
• Better code documentation
• Even better performance