- Execute
docker-compose up
in your terminal. - Open the jupyter notebook in your browser.
Use a recurrent neural network to produce new fortune cookie fortunes. This repo shows three different attempts to produce a working model to produce new fortunes.
- Attempt One: GRU with corpus embedding. Largely based on this blog post. TODO: Add blog post link.
- Attempt Two: Pretrained embedding layer using glove.
- Attempt Three: Fine tune GPT-2 within a docker container locally.
- It has been difficult to find the volume of data required to train a natrual language processing model on fortune cookies. There are not very many fortune cookie databases out in the wild.
- Fortune cookies are built from somewhat atypical English, which has made the lack of data more difficult.