This project is a proof of concept for using textgenrnn to generate terminal story text for [Bungie's Marathon](https://en.wikipedia.org/wiki/Marathon_(video_game\)) game.
This is intended to not only show off the possibilities of training a model to create background stories but also show off some of the features MissingLink.ai provides when training models locally and in the cloud.
Here is an example of some text that was generated by a trained model:
* * * incoming message * * *
The Marathon is equipped with three artificial intelligence units. Tycho handles the science and engineering network, Durandal controls the marathon autonomous functions such as doors, air from kitchens, and I don't remember what to do.
The aliens seem to have been caught off guard by the strength of our counterattack. This is good news, but I have detected more ships landing on the marathon, and I fear that the aliens are reinforcing their efforts.
* * * end of message * * *
To run this project, use the train.py script to create a new model and the generate.py script to output sample text. There is a dataset comprised of Marathon's terminal text scraped from the official story site.
You can use the requirements.txt file to install the project's dependencies. There is also a config.py script you can use to modify settings during training and text generation.
Max Woolf (@minimaxir) - textgenrnn
Max's open-source projects are supported by his Patreon. If you found this project helpful, any monetary contributions to the Patreon are appreciated and will be put to good creative use.
Andrej Karpathy for the original proposal of the char-rnn via the blog post The Unreasonable Effectiveness of Recurrent Neural Networks.
Daniel Grijalva for contributing an interactive mode.