I wanted to create a simple recipe generator using a recurrent neural network. I chose Max Woolf's implementation of textgenrnn as my language model due to it's ease of use and flexibility. In designing this project I aimed more to generate humurous yet understandable recipes.
I ended up deciding to train two seperate models:
- A title generator
- An instruction generator
The original plan was to try to integrate some information from the generated title to the instruction generator for context, but I have yet to find an easy way to do this with the current model.
I gathered my data using the Spoonacular API and the Spoonacular Python wrapper. Unfortunately, the wrapper stopped working halfway through this project, so I did not end up collecting as much data as I would have liked. I have left the script to gather the recipes for documentation purposes.