Repository of the materials for my Web Dev & Sausages, vol. 7 talk in May 17th 2018. Session title is "Deep Learning Lyrics" and is about machine generating lyrics based on existing lyrics with Keras and Tensorflow.
Contains environment.yml-file to use with Anaconda "conda env create". Download Anaconda from https://www.anaconda.com/download/
Contains file tensorflow-1.8.0-cp36-cp36m-win_amd64.whl which you can also use if you have a hard time finding a good build from Tensorflow (that supports all available CPU instructions)
Start training process by running
python main.py ../data/cannibalcorpse.txt 400
Once you have trained the model and are ready to generate more lyrics than during the training, run:
python lyrgen.py cannibalcorpse.h5 ../data/cannibalcorpse.txt 1.2
Last parameter is the diversity (temperature). Higher value means more errors but maybe more "innovative" results.
- Keras - Keras is a high-level neural networks API, written in Python and capable of running on top of TensorFlow, CNTK, or Theano
- TensorFlow - TensorFlow is an open source software library for high performance numerical computation
- Antti Simonen - Initial work - antsim
See also the list of contributors who participated in this project.
This project is licensed under the MIT License - see the LICENSE.md file for details
- Hat tip to anyone who's code was used, especially Ivan Liljeqvist - ivan-liljeqvist
- Gofore Hackathon team who originally played around with this.