GPT Storytale

Create controllable story sequences with transformers that can be used in user-evaluation studies of text-to-visualisation pipelines.

Getting started

GPT Storytale has been tested on python 3.10. You will need to install the required packages. Using a virtual environment is recommended. e.g.

python3 -m venv myvenv          # set up the virtual env
source myvenv/bin/activate      # activate the virtual env (Linux) (On Windows: myvenv\Scripts\activate)
pip install -r requirements.txt # install the required packages

You will also need to set up your OpenAI API key. You can do this by creating a .env file in the root directory of the project and adding the following line:

OPENAI_API_KEY=your-api-key

(see .env-template for an example)

Usage

To run the GPT Storytale system:

1. Generate a story sequence:

python src/generate_story.py --input your_input.json --output output_folder

2. Evaluate the generated sequences:

python src/evaluate.py --story_path output_folder

For detailed examples and parameter options, see the examples directory.

Project Structure

GPT Storytale/
├── src/                # Source code
├── examples/           # Example inputs and outputs
├── tests/             # Test files
├── figs/              # Figures and diagrams
└── requirements.txt   # Project dependencies

Features

• Generate coherent story sequences using transformer models
• Controllable narrative parameters
• Support for multiple story genres and styles
• Evaluation metrics for generated sequences
• Easy integration with visualization pipelines

Acknowledgements

This project builds upon several open-source tools and libraries:

• OpenAI GPT - For text generation
• Transformers - Hugging Face's transformer models
• NumPy - For numerical computations
• Pandas - For data manipulation

Contributing

We welcome contributions! Please follow these steps:

1. Fork the repository
2. Create a new branch for your feature
3. Submit a pull request with a clear description of your changes

License

This project is licensed under the MIT License - see the LICENSE file for details.