Create new text via machine learning of existing source material. Uses Recurrent Neural Networks (RNN) using textgenrnn, tensorflow libraries.
Built in Python, using the textgenrrn library (tensorflow).
See the milestones page for work-in-progress and planned features.
Also see CHANGELOG.md for project history.
There are Windows binaries available now. This is by far the easiest way to run simple-text-generator. Created with PyInstaller, it includes all dependencies including Python.
Best way to learn ML is by trying it out. Fire up the UI, throw some data at it, and see the results.
Also check the wiki for additional information and known issues.
The following instructions are examples; adapt them for your needs.
Requires:
- <= Python 3.8 (not sure earliest version) due to Tensorflow incompatibilities with Python 3.9+ right now.
- Recommend using pyenv to install latest 3.8
- pipenv See https://pypi.org/project/pipenv/
- Optional Tensorflow CUDA libraries for GPU training. CPU training is used otherwise. Be warned that setup is not straightforward and I currently do not have instructions.
git clone [email protected]:thtroyer/simple-text-generator.git
cd simple-text-generator
pipenv install Pipfile
# run the executable code
python simple-text-generator-iu.py
python run_training.py
Tagged releases will be the most stable.
Features will often be created as feature branches and merged into master as they are completed. master branch will be the latest code and should be in mostly-working order. Use this if you want the latest, but possibly less stable than the released version.
This project uses semantic versioning.
As a developer primarily familiar with Java and PHP, this is currently my playground for learning Python and ML, which includes making mistakes and messes, and then figuring out improvements.
That said, I welcome contributions, bugfixes, and feedback for anyone interested in helping move this project along. Try to keep any pull requests small and focused for easy review and acceptance. Overly large changes will be slow to accept or rejected.
This code, especially in its current state, should not be used as reference, study, or as a measure of my typical code quality. I'm currently aiming for functionality and learning, which requires some mucking around. I hope to be able to take this disclaimer off soon as I become more familiar with Python idioms, architecture, syntax, and add some unit tests.
Until then, read the code at your own risk. Here be dragons.