Cuisine Transformation

By swapping out ingredients, we aim to transform one cuisine to another. This is a small subproblem to the question Can Computers Cook .


Data is a property of Complex Systems Lab IIITD hence not available in the repo.
Word2Vec used to vectorize recipes in the Encoder-Decoder model.
Scoring Scheme - Output Ingredient gets +1 if its category matches with Input Ingredient. So the score for the recipe becomes the average of the individual score and the final score is the average of recipe scores.
To tackle the cuisine transformation problem, two different models were used.

Drawbacks/ Potential Issues - This problem is a subproblem os the AI recipe generation problem and it would be better if it was a tackle with that perspective rather than considering it an NLP problem. If ingredient flavor molecules could be known and quantified, they would possibly act as great features for such a problem. Also, when the previous issues are dealt with, issues with the scoring criteria would also be solved because currently for a prediction to be correct an additive must always be replaced by an additive and so on for other categories.