Marco

A Markov-chains based text generator (currently a work in progress).

Currently at version v0.2.1.0. Click here to read the changelog.

Build

We recommend to get and use stack in order to build, run and install the project.

With stack

stack build

You can run the program with stack run, or install it into your setup with stack install and then run it with marco.

With GHC

In case you don't have stack (note that this is not supported and might not work).

cd src
ghc Main.hs -o marco

You will find the marco executable in src/.

Usage

Assuming the program has been built properly and the executable is called marco (replace marco with stack run if necessary):

Training the algorithm

marco train <foldername> <context_length>

Where

• foldername is the (absolute or relative) path to a folder with the files to be used as the base texts
• context_length is the amount of words to be considered as a "unit" in order to find the next suffix

For training to be successful, you need to save one or several .txt files within the specified folder (we suggest using the data folder in the repository).

Generating text

marco generate <max_length>

Where

• max_length is an upper bound for the generated text length (i.e. an approximate, not necessarily exact, max length). In practice, the text's length may vary from 0 (which would be a really bad scenario that might imply that your base texts are not adecuate for the task) to (max_length + context_length).

To-do list

What follows is a non-exhaustive list of possible features to be added to marco:

• Improved performance, particularly in training
• An easy-to-use API to integrate text generation into other programs (e.g. bots)
• Better normalization algorithms
• The ability to define the first token to be used in text generation
• Code refactoring
• Exception handling
• Edge-case testing
• Sample texts: inputs and outputs
• Improved README: add description, motivation, etc!