Food vectors. Live demo at https://altosaar.github.io/food2vec/, blog post with more information and plots here: https://jaan.io/food2vec-augmented-cooking-machine-intelligence/
Train a model on the recipes dataset, replicate the results from the blog post:
conda env create -f environment.yml
conda activate food2vec
git clone [email protected]:altosaar/food2vec.git
echo "[submodule \"src/sentence_word2vec\"]
path = src/sentence_word2vec
url = https://github.com/altosaar/sentence_word2vec.git
git submodule update --init
cd food2vec/src
./run_fasttext.sh
# run t-sne and make the plots for the ingredient embeddings
jupyter notebook ./src/plot_ingredients_recipes.ipynb
https://gist.github.com/altosaar/67d8456ad28acd1abb497f1950d8de8a
Pull requests and all feedback welcome! Please file an issue if you run into problems replicating the results.
- get more data
- convert jupyter notebook for plotting into one python script
- write scripts to figure out the right vocabulary
- fit a better model (e.g. multi-class regression in pytorch) -- if you manage to get better results than the live demo at https://altosaar.github.io/food2vec/ just submit a pull request with the new
assets/data/wordVecs.jsand I'll happily update it :) - compare the above model embeddings to the current embeddings
- make the UI of the website more user-friendly and mobile-friendly
Thanks to Anthony for open-sourcing a javascript embedding browser -- the one here is heavily based on it.